diff --git a/CHANGELOG.md b/CHANGELOG.md
index 8990028d..6916814d 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,36 @@
 
 ## [Unreleased]
 
+## [v0.50.255] — 2026-05-01
+
+### Added
+- **Insights panel — usage analytics dashboard** (#464) — new `GET /api/insights?days=N` endpoint walks `_index.json` (no full session loads) and aggregates session/message/token counts, model breakdown, and activity-by-day-of-week + activity-by-hour. New nav rail entry between Todos and Settings; the panel renders stats cards, a token breakdown row, and ASCII-style horizontal-bar charts. Period filter (7/30/90 days). (`api/routes.py`, `static/panels.js`, `static/index.html`, `static/i18n.js`, `static/style.css`) @bergeouss — PR #1405, fixes #464
+
+- **Rollback UI — restore from agent checkpoints** (#466) — new `api/rollback.py` exposes 3 endpoints (`GET /api/rollback/list`, `GET /api/rollback/diff`, `POST /api/rollback/restore`) over the agent's `CheckpointManager` shadow git repos at `{hermes_home}/checkpoints/<sha256-of-canonical-workspace>/<commit_hash>/.git`. Workspace is allowlisted via `load_workspaces()` (added during contributor security pass `d9f3a69`). `_validate_checkpoint_id()` regex-guards the checkpoint parameter against path-traversal (Opus pre-release advisor finding — `Path()` does NOT normalize `..`). Restore copies files via `shutil.copy2` and never deletes; diff uses `difflib.unified_diff`. (`api/rollback.py`, `api/routes.py`) @bergeouss — PR #1405, fixes #466
+
+- **Turn-based voice mode — STT + TTS chained flow** — new voice-mode button in the composer; activating it puts the agent in a listen → send → think → speak → listen loop. Uses the browser's Web Speech API (gated on both `SpeechRecognition` AND `speechSynthesis` support). Auto-send on 1.8s silence after a final transcript. Honors saved voice preferences (`hermes-tts-voice`, `hermes-tts-rate`, `hermes-tts-pitch`). Bails out on `not-allowed` / `service-not-allowed` / `audio-capture` errors. **Pre-release fix:** the patched `autoReadLastAssistant` fired globally — if the user navigated to a different session between send and stream completion, TTS would speak the wrong session's reply. Now captures `S.session.session_id` at thinking-time and bails to listening if the active session changed. (Opus pre-release advisor.) (`static/boot.js`, `static/i18n.js`, `static/index.html`, `static/style.css`) @bergeouss — PR #1405
+
+- **API redact toggle — opt out of response-layer redaction** — adds `api_redact_enabled` setting (defaults to `True` so existing users see no behavioral change). When disabled, `redact_session_data()` returns payloads as-is. Useful for users who pipe the WebUI API into automation that needs the original strings. (`api/helpers.py`, `api/config.py`, `static/panels.js`, `static/i18n.js`) @bergeouss — PR #1405
+
+- **Subagent tree visualization** — UI affordance for sessions that spawn subagents. (`static/panels.js`, `static/sessions.js`, `static/style.css`, `static/i18n.js`) @bergeouss — PR #1405
+
+### Fixed
+- **Session provider context preserved across model picker → runtime resolution** (#1240) — the WebUI model picker can show multiple providers exposing the same bare model id (e.g. `gpt-5.5` from OpenAI Codex, OpenRouter, Copilot). Previously sessions persisted only the bare model, so a session selected as "gpt-5.5 from OpenAI Codex" silently rerouted through whatever provider became default after a config change. New `model_provider: str | None` field on `Session` is persisted in metadata, threaded through every chat path (`/api/session/new`, `/api/session/update`, `/api/chat/start`, `/api/chat/sync`, `/btw`, `/background`, `_run_agent_streaming`), and is gated in `compact()` to emit only when truthy (matches v0.50.251 lineage end_reason gating). New `model_with_provider_context(model_id, model_provider)` in `api/config.py` builds the `@provider:model` form when provider differs from configured default, then passes through `resolve_model_provider()`. New `_should_attach_codex_provider_context()` narrow exception detects bare GPT-* models under active OpenAI Codex (because Codex/OpenRouter/Copilot expose overlapping GPT names). New `_resolve_compatible_session_model_state()` returns `(effective_model, effective_provider, model_was_normalized)`. Frontend adds `MODEL_STATE_KEY='hermes-webui-model-state'` localStorage with structured persistence and migrates from the legacy `hermes-webui-model` key. 13 new tests in `test_provider_mismatch.py`, 2 in `test_model_picker_badges.py`. (`api/config.py`, `api/models.py`, `api/routes.py`, `api/streaming.py`, `static/boot.js`, `static/messages.js`, `static/panels.js`, `static/sessions.js`, `static/ui.js`) @starship-s — PR #1390, refs #1240
+
+- **TTS toggle: speaker icon never appeared when "Text-to-Speech for responses" was ticked** (#1409, closes #1409) — `_applyTtsEnabled()` set `btn.style.display=enabled?'':'none'` on every `.msg-tts-btn`. The `''` branch removes the inline override, after which the `.msg-tts-btn{display:none;}` rule from `style.css` re-hides the button. Both the "enabled" and "disabled" branches left the icon hidden, so the toggle had no visible effect since the feature shipped in #499. Fixed by switching to a body-class toggle (`body.tts-enabled`) plus a compound CSS selector (`body.tts-enabled .msg-tts-btn{display:inline-flex;}`). The new shape bypasses the `.msg-action-btn` / `.msg-tts-btn` cascade collision and survives subsequent `renderMd()` re-renders without re-querying every button. (`static/panels.js`, `static/style.css`, `tests/test_499_tts_playback.py`) — PR #1411, fixes #1409, reported by @AvidFuturist via Discord
+
+- **Ollama (local) no longer falsely reports "API key configured" when only Ollama Cloud key is set** (#1410, closes #1410) — both providers were mapped to the same `OLLAMA_API_KEY` env var in `_PROVIDER_ENV_VAR`, so configuring Ollama Cloud lit up the local Ollama card too. The runtime in `hermes_cli/runtime_provider.py` only consumes `OLLAMA_API_KEY` when the base URL hostname is `ollama.com` — local Ollama is keyless by design — so the WebUI was reporting "configured" for a key local Ollama doesn't even read. Dropped the bare `"ollama": "OLLAMA_API_KEY"` mapping; local Ollama users who genuinely need a key can still set `providers.ollama.api_key` in `config.yaml`, and `_provider_has_key()` continues to honor that path. (`api/providers.py`, `tests/test_provider_management.py`) — PR #1411, fixes #1410, reported by @AvidFuturist via Discord
+
+### Changed
+
+- **`api/rollback.py` — checkpoint id regex validation (defense-in-depth)** — Opus pre-release follow-up. The `checkpoint` parameter on `/api/rollback/diff` and `/api/rollback/restore` was joined into the path via `_checkpoint_root() / ws_hash / checkpoint`. `Path("/a/b") / "../escape"` does NOT normalize, so an authenticated caller could pass `../<other-ws-hash>/<sha>` and read or restore from another allowlisted workspace's checkpoint store. New `_validate_checkpoint_id()` regex-guards with `^[A-Za-z0-9_-][A-Za-z0-9_.-]{0,63}$` and rejects literal `.` / `..`. (`api/rollback.py`)
+
+- **`redact_session_data()` reads `api_redact_enabled` once per response, not per string** — Opus pre-release follow-up. The new `_redact_text` per-string `load_settings()` call (added by #1405's redact-toggle feature) caused hundreds of disk reads + JSON parses per `/api/session?session_id=X` response on a 50-message session — every nested string in `messages[]` and `tool_calls[]` recursed back into `_redact_value` → `_redact_text` → `load_settings`. Now read once at the top of `redact_session_data()` and threaded through via a private `_enabled` keyword. Fast path when disabled: still walks but returns immediately. (`api/helpers.py`, `tests/test_v050255_opus_followups.py`)
+
+- **Voice mode pins active session id at thinking-time** — Opus pre-release follow-up. The patched `autoReadLastAssistant` fires globally; if the user navigated to a different session between sending a turn and stream completion, TTS would speak the wrong session's last assistant message. New `_voiceModeThinkingSid` closure variable captures `S.session.session_id` in `_voiceModeSend`; `_speakResponse` bails to `_startListening()` if the current sid no longer matches. (`static/boot.js`, `tests/test_v050255_opus_followups.py`)
+
+- **`api/rollback.py::_inspect_checkpoint` drops bare `Exception` from except tuple** — Opus pre-release follow-up. The previous `except (subprocess.TimeoutExpired, OSError, Exception)` made the specific catches redundant and swallowed everything. Now `(subprocess.TimeoutExpired, OSError)` only. (`api/rollback.py`, `tests/test_v050255_opus_followups.py`)
+
 ## [v0.50.254] — 2026-05-01
 
 ### Fixed
diff --git a/api/config.py b/api/config.py
index 90b788a6..4e7f8ad5 100644
--- a/api/config.py
+++ b/api/config.py
@@ -1034,6 +1034,20 @@ def resolve_model_provider(model_id: str) -> tuple:
         _PORTAL_PROVIDERS = {"nous", "opencode-zen", "opencode-go", "nvidia"}
         if config_provider in _PORTAL_PROVIDERS:
             return model_id, config_provider, config_base_url
+        # The OpenAI Codex provider uses a real base_url, but its default
+        # ChatGPT endpoint cannot serve OpenRouter-style provider/model IDs.
+        # Keep that narrow exception before the custom endpoint protection so
+        # selecting openai/gpt-5.5 from OpenRouter under active Codex still
+        # routes through OpenRouter. Other base_url-backed real providers may be
+        # custom/proxy endpoints, so they must fall through to the branch below.
+        if (
+            config_provider == "openai-codex"
+            and str(config_base_url or "").strip().rstrip("/")
+            == "https://chatgpt.com/backend-api/codex"
+            and prefix in _PROVIDER_MODELS
+            and prefix != config_provider
+        ):
+            return model_id, "openrouter", None
         # If a custom endpoint base_url is configured, don't reroute through OpenRouter
         # just because the model name contains a slash (e.g. google/gemma-4-26b-a4b).
         # The user has explicitly pointed at a base_url, so trust their routing config.
@@ -1056,6 +1070,42 @@ def resolve_model_provider(model_id: str) -> tuple:
     return model_id, config_provider, config_base_url
 
 
+def model_with_provider_context(model_id: str, model_provider: str | None = None) -> str:
+    """Return the model string to pass to ``resolve_model_provider()``.
+
+    Session persistence keeps the user's selected provider in ``model_provider``
+    instead of forcing every selected model into ``@provider:model`` form. At
+    runtime, however, ``resolve_model_provider()`` still understands that
+    internal disambiguation form, so use it only when the provider context is
+    needed to route away from the current default provider.
+    """
+    model = str(model_id or "").strip()
+    provider = str(model_provider or "").strip().lower()
+    if not model or not provider or provider == "default" or model.startswith("@"):
+        return model
+
+    model_cfg = cfg.get("model", {})
+    config_provider = None
+    if isinstance(model_cfg, dict):
+        config_provider = str(model_cfg.get("provider") or "").strip().lower()
+
+    # If the selected provider is already the configured provider, leaving the
+    # model bare preserves provider-specific base_url/proxy settings.
+    if provider == config_provider:
+        return model
+
+    # OpenRouter selections with slash IDs are explicit provider/model paths.
+    if provider == "openrouter":
+        return f"@{provider}:{model}"
+
+    # For non-OpenRouter slash IDs, keep the ID intact so existing custom/proxy
+    # base_url routing and portal-provider handling remain in charge.
+    if "/" in model:
+        return model
+
+    return f"@{provider}:{model}"
+
+
 def get_effective_default_model(config_data: dict | None = None) -> str:
     """Resolve the effective Hermes default model from config, then env overrides."""
     active_cfg = config_data if config_data is not None else cfg
@@ -2232,6 +2282,7 @@ _SETTINGS_DEFAULTS = {
     "notifications_enabled": False,  # browser notification when tab is in background
     "show_thinking": True,  # show/hide thinking/reasoning blocks in chat view
     "simplified_tool_calling": True,  # group tools/thinking into one quiet activity disclosure
+    "api_redact_enabled": True,  # redact sensitive data (API keys, secrets) from API responses
     "sidebar_density": "compact",  # compact | detailed
     "auto_title_refresh_every": "0",  # adaptive title refresh: 0=off, 5/10/20=every N exchanges
     "busy_input_mode": "queue",  # behavior when sending while agent is running: queue | interrupt | steer
@@ -2349,6 +2400,7 @@ _SETTINGS_BOOL_KEYS = {
     "notifications_enabled",
     "show_thinking",
     "simplified_tool_calling",
+    "api_redact_enabled",
 }
 # Language codes are validated as short alphanumeric BCP-47-like tags (e.g. 'en', 'zh', 'fr')
 _SETTINGS_LANG_RE = __import__("re").compile(r"^[a-zA-Z]{2,10}(-[a-zA-Z0-9]{2,8})?$")
diff --git a/api/helpers.py b/api/helpers.py
index 50574eba..f6c8b584 100644
--- a/api/helpers.py
+++ b/api/helpers.py
@@ -175,17 +175,39 @@ def _build_redact_fn():
     return _combined_redact
 
 
-_redact_text = _build_redact_fn()
+_redact_fn_cached = _build_redact_fn()
 
 
-def _redact_value(v):
-    """Recursively redact credentials from strings, dicts, and lists."""
+def _redact_text(text: str, *, _enabled: bool | None = None) -> str:
+    """Redact sensitive text from API responses. Respects api_redact_enabled setting.
+
+    The ``_enabled`` parameter is an internal optimization for callers that
+    redact many strings in a single response — `redact_session_data()` reads
+    the setting once and threads it through ``_redact_value`` so we avoid
+    re-loading settings.json from disk per string. (Opus pre-release perf fix.)
+    """
+    if not isinstance(text, str) or not text:
+        return text
+    if _enabled is None:
+        from api.config import load_settings
+        _enabled = bool(load_settings().get("api_redact_enabled", True))
+    if not _enabled:
+        return text
+    return _redact_fn_cached(text)
+
+
+def _redact_value(v, *, _enabled: bool | None = None):
+    """Recursively redact credentials from strings, dicts, and lists.
+
+    ``_enabled`` is threaded through so a single response-level redact pass
+    only reads settings.json once. (Opus pre-release perf fix.)
+    """
     if isinstance(v, str):
-        return _redact_text(v)
+        return _redact_text(v, _enabled=_enabled)
     if isinstance(v, dict):
-        return {k: _redact_value(val) for k, val in v.items()}
+        return {k: _redact_value(val, _enabled=_enabled) for k, val in v.items()}
     if isinstance(v, list):
-        return [_redact_value(item) for item in v]
+        return [_redact_value(item, _enabled=_enabled) for item in v]
     return v
 
 
@@ -194,14 +216,22 @@ def redact_session_data(session_dict: dict) -> dict:
 
     Applies to: messages[], tool_calls[], and title.
     The underlying session file is not modified; redaction is response-layer only.
+
+    Reads the ``api_redact_enabled`` setting ONCE for the entire response and
+    threads it through to avoid hundreds of settings.json reads per session
+    payload (a 50-message session has hundreds of nested strings). When the
+    setting is disabled this is also a fast path: the recursion still walks
+    but every string returns early.
     """
+    from api.config import load_settings
+    _enabled = bool(load_settings().get("api_redact_enabled", True))
     result = dict(session_dict)
     if isinstance(result.get('title'), str):
-        result['title'] = _redact_text(result['title'])
+        result['title'] = _redact_text(result['title'], _enabled=_enabled)
     if 'messages' in result:
-        result['messages'] = _redact_value(result['messages'])
+        result['messages'] = _redact_value(result['messages'], _enabled=_enabled)
     if 'tool_calls' in result:
-        result['tool_calls'] = _redact_value(result['tool_calls'])
+        result['tool_calls'] = _redact_value(result['tool_calls'], _enabled=_enabled)
     return result
 
 
diff --git a/api/models.py b/api/models.py
index 6ab78174..2a1f9c19 100644
--- a/api/models.py
+++ b/api/models.py
@@ -306,6 +306,7 @@ def _lookup_index_message_count(session_id):
 class Session:
     def __init__(self, session_id: str=None, title: str='Untitled',
                  workspace=str(DEFAULT_WORKSPACE), model=DEFAULT_MODEL,
+                 model_provider=None,
                  messages=None, created_at=None, updated_at=None,
                  tool_calls=None, pinned: bool=False, archived: bool=False,
                  project_id: str=None, profile=None,
@@ -326,6 +327,7 @@ class Session:
         self.title = title
         self.workspace = str(Path(workspace).expanduser().resolve())
         self.model = model
+        self.model_provider = str(model_provider).strip().lower() if model_provider else None
         self.messages = messages or []
         self.tool_calls = tool_calls or []
         self.created_at = created_at or time.time()
@@ -366,7 +368,7 @@ class Session:
         # without parsing the full messages array (which may be 400KB+).
         # Fields are listed in the order they should appear in the JSON file.
         METADATA_FIELDS = [
-            'session_id', 'title', 'workspace', 'model', 'created_at', 'updated_at',
+            'session_id', 'title', 'workspace', 'model', 'model_provider', 'created_at', 'updated_at',
             'pinned', 'archived', 'project_id', 'profile',
             'input_tokens', 'output_tokens', 'estimated_cost',
             'personality', 'active_stream_id',
@@ -448,6 +450,7 @@ class Session:
             'title': self.title,
             'workspace': self.workspace,
             'model': self.model,
+            'model_provider': self.model_provider,
             'message_count': (
                 self._metadata_message_count
                 if self._metadata_message_count is not None
@@ -702,7 +705,7 @@ def get_session(sid, metadata_only=False):
         return s
     raise KeyError(sid)
 
-def new_session(workspace=None, model=None, profile=None):
+def new_session(workspace=None, model=None, profile=None, model_provider=None):
     """Create a new in-memory session.
 
     The session lives in the SESSIONS dict only — no disk write happens until
@@ -736,6 +739,7 @@ def new_session(workspace=None, model=None, profile=None):
     s = Session(
         workspace=workspace or get_last_workspace(),
         model=effective_model,
+        model_provider=model_provider,
         profile=profile,
     )
     with LOCK:
diff --git a/api/providers.py b/api/providers.py
index a36a935b..424a7927 100644
--- a/api/providers.py
+++ b/api/providers.py
@@ -44,7 +44,14 @@ _PROVIDER_ENV_VAR: dict[str, str] = {
     "x-ai": "XAI_API_KEY",
     "opencode-zen": "OPENCODE_ZEN_API_KEY",
     "opencode-go": "OPENCODE_GO_API_KEY",
-    "ollama": "OLLAMA_API_KEY",
+    # NOTE: bare "ollama" (local) deliberately omitted — local Ollama is keyless
+    # by default and the runtime in hermes_cli/runtime_provider.py only consumes
+    # OLLAMA_API_KEY when the base URL hostname is ollama.com (Ollama Cloud).
+    # If we mapped both providers to the same env var, configuring Ollama Cloud
+    # would falsely flip the local Ollama card to "API key configured" (#1410).
+    # Users who genuinely run an authenticated local Ollama can still set a key
+    # via providers.ollama.api_key in config.yaml — that path remains supported
+    # by _provider_has_key().
     "ollama-cloud": "OLLAMA_API_KEY",
     "nvidia": "NVIDIA_API_KEY",
 }
diff --git a/api/rollback.py b/api/rollback.py
new file mode 100644
index 00000000..18d888fd
--- /dev/null
+++ b/api/rollback.py
@@ -0,0 +1,320 @@
+"""
+Hermes Web UI -- Filesystem checkpoint (rollback) API.
+
+Provides endpoints to list, diff, and restore filesystem checkpoints
+created by the Hermes agent's CheckpointManager.  Checkpoints live at
+``{hermes_home}/checkpoints/<hash>/`` as shadow git repositories.
+"""
+
+import hashlib
+import json
+import logging
+import os
+import re
+import shutil
+import subprocess
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any
+
+logger = logging.getLogger(__name__)
+
+# Checkpoint identifiers are SHA-style hex hashes from the agent's
+# CheckpointManager. We only allow [A-Za-z0-9_.-]{1,64} (no '/' so the
+# value cannot be a path separator, no leading '.' so it cannot escape
+# upward via '..'/'.'). This is defense-in-depth: the workspace arg is
+# already allowlisted, but ``Path() / "../escape"`` does not normalize,
+# so without this guard a `checkpoint` value of `../<other-ws-hash>/<sha>`
+# would let any authenticated caller diff or restore from another
+# allowlisted workspace's checkpoint store. (Opus pre-release advisor.)
+_CHECKPOINT_ID_RE = re.compile(r"^[A-Za-z0-9_-][A-Za-z0-9_.-]{0,63}$")
+
+
+def _validate_checkpoint_id(checkpoint: str) -> str:
+    cid = str(checkpoint or "").strip()
+    if not cid or cid in (".", "..") or not _CHECKPOINT_ID_RE.fullmatch(cid):
+        raise ValueError(
+            "checkpoint id must match [A-Za-z0-9_-][A-Za-z0-9_.-]{0,63}"
+        )
+    return cid
+
+
+def _hermes_home() -> Path:
+    """Return the active Hermes home directory."""
+    try:
+        from api.profiles import get_active_hermes_home
+        return Path(get_active_hermes_home())
+    except Exception:
+        return Path(os.environ.get("HERMES_HOME", "~/.hermes")).expanduser()
+
+
+def _workspace_hash(workspace: str) -> str:
+    """Derive the checkpoint directory name from a workspace path.
+
+    Matches the agent's CheckpointManager._get_checkpoint_dir logic:
+    SHA-256 of the canonical workspace path.
+    """
+    try:
+        canonical = os.path.realpath(workspace)
+    except (OSError, ValueError):
+        canonical = workspace
+    return hashlib.sha256(canonical.encode()).hexdigest()[:12]
+
+
+def _checkpoint_root() -> Path:
+    return _hermes_home() / "checkpoints"
+
+
+def _resolve_workspace(workspace: str) -> str:
+    """Validate and return the canonical workspace path.
+
+    Security: workspace must match a known configured workspace
+    (from workspaces.json or session-attached workspaces).
+    """
+    if not workspace or not isinstance(workspace, str):
+        raise ValueError("workspace is required")
+    # Basic path validation
+    resolved = os.path.realpath(workspace)
+    if not os.path.isdir(resolved):
+        raise ValueError(f"Workspace does not exist: {workspace}")
+    # Security: confirm workspace is in the known list
+    try:
+        from api.workspace import load_workspaces
+        known_paths = set()
+        for ws in load_workspaces():
+            p = ws.get("path", "")
+            if p:
+                known_paths.add(os.path.realpath(p))
+        if resolved not in known_paths:
+            raise ValueError(f"Workspace not in configured list: {workspace}")
+    except ImportError:
+        logger.warning("Could not load workspace list for rollback validation")
+    return resolved
+
+
+def _find_git() -> str:
+    """Return the path to the git binary."""
+    return shutil.which("git") or "git"
+
+
+# ── Public API functions (called from routes.py) ────────────────────────────
+
+
+def list_checkpoints(workspace: str) -> dict[str, Any]:
+    """List all checkpoints for a workspace.
+
+    Returns a dict with:
+        checkpoints: list of checkpoint objects
+        workspace: resolved workspace path
+        checkpoint_dir: the checkpoint directory path
+    """
+    resolved = _resolve_workspace(workspace)
+    ws_hash = _workspace_hash(resolved)
+    ckpt_dir = _checkpoint_root() / ws_hash
+
+    checkpoints = []
+    if not ckpt_dir.is_dir():
+        return {"checkpoints": [], "workspace": resolved, "checkpoint_dir": str(ckpt_dir)}
+
+    # Each checkpoint is a git repo in <ckpt_dir>/<commit_hash>/
+    git = _find_git()
+    for entry in sorted(ckpt_dir.iterdir(), key=lambda p: p.stat().st_mtime if p.is_dir() else 0, reverse=True):
+        if not entry.is_dir():
+            continue
+        ckpt_info = _inspect_checkpoint(entry, git)
+        if ckpt_info:
+            checkpoints.append(ckpt_info)
+
+    return {
+        "checkpoints": checkpoints,
+        "workspace": resolved,
+        "checkpoint_dir": str(ckpt_dir),
+    }
+
+
+def _inspect_checkpoint(ckpt_path: Path, git: str) -> dict[str, Any] | None:
+    """Extract metadata from a single checkpoint directory."""
+    git_dir = ckpt_path / ".git"
+    if not git_dir.is_dir():
+        return None
+
+    name = ckpt_path.name
+    try:
+        result = subprocess.run(
+            [git, "-C", str(ckpt_path), "log", "--format=%H%n%s%n%aI", "-1"],
+            capture_output=True, text=True, timeout=5,
+        )
+        if result.returncode != 0 or not result.stdout.strip():
+            return None
+
+        lines = result.stdout.strip().split("\n")
+        commit_hash = lines[0] if len(lines) > 0 else name
+        message = lines[1] if len(lines) > 1 else "checkpoint"
+        date_str = lines[2] if len(lines) > 2 else ""
+
+        # Parse date for display
+        date_display = ""
+        if date_str:
+            try:
+                dt = datetime.fromisoformat(date_str)
+                date_display = dt.strftime("%Y-%m-%d %H:%M")
+            except (ValueError, TypeError):
+                date_display = date_str
+
+        # Count files
+        files_result = subprocess.run(
+            [git, "-C", str(ckpt_path), "ls-files"],
+            capture_output=True, text=True, timeout=5,
+        )
+        file_count = len(files_result.stdout.strip().split("\n")) if files_result.stdout.strip() else 0
+
+        return {
+            "id": name,
+            "commit": commit_hash[:12],
+            "message": message,
+            "date": date_str,
+            "date_display": date_display,
+            "files": file_count,
+            "path": str(ckpt_path),
+        }
+    except (subprocess.TimeoutExpired, OSError) as e:
+        logger.debug("Failed to inspect checkpoint %s: %s", ckpt_path, e)
+        return None
+
+
+def get_checkpoint_diff(workspace: str, checkpoint: str) -> dict[str, Any]:
+    """Show the diff between a checkpoint and the current workspace state.
+
+    Returns a dict with:
+        diff: unified diff text
+        files_changed: list of changed file paths
+    """
+    resolved = _resolve_workspace(workspace)
+    checkpoint = _validate_checkpoint_id(checkpoint)
+    ws_hash = _workspace_hash(resolved)
+    ckpt_dir = _checkpoint_root() / ws_hash / checkpoint
+
+    if not ckpt_dir.is_dir():
+        raise ValueError(f"Checkpoint not found: {checkpoint}")
+
+    git = _find_git()
+
+    # Get list of files in the checkpoint
+    ls_result = subprocess.run(
+        [git, "-C", str(ckpt_dir), "ls-files"],
+        capture_output=True, text=True, timeout=10,
+    )
+    if ls_result.returncode != 0:
+        raise ValueError("Failed to list checkpoint files")
+
+    ckpt_files = [f for f in ls_result.stdout.strip().split("\n") if f]
+    files_changed = []
+    diff_lines = []
+
+    for rel_path in ckpt_files:
+        ckpt_file = ckpt_dir / rel_path
+        ws_file = Path(resolved) / rel_path
+
+        if not ckpt_file.is_file():
+            continue
+
+        # Read checkpoint version
+        try:
+            ckpt_content = ckpt_file.read_text(errors="replace")
+        except OSError:
+            continue
+
+        # Read workspace version (if exists)
+        if ws_file.is_file():
+            try:
+                ws_content = ws_file.read_text(errors="replace")
+            except OSError:
+                ws_content = ""
+        else:
+            ws_content = None  # File was deleted in workspace
+
+        if ws_content is None:
+            # File exists in checkpoint but not in workspace (deleted)
+            files_changed.append({"file": rel_path, "status": "deleted"})
+            diff_lines.append(f"--- a/{rel_path}")
+            diff_lines.append(f"+++ /dev/null")
+            diff_lines.append("@@ -1,{lines} +0,0 @@".format(lines=len(ckpt_content.splitlines())))
+            for line in ckpt_content.splitlines():
+                diff_lines.append(f"-{line}")
+        elif ckpt_content != ws_content:
+            # File changed
+            import difflib
+            ckpt_lines = ckpt_content.splitlines(keepends=True)
+            ws_lines = ws_content.splitlines(keepends=True)
+            diff = list(difflib.unified_diff(ckpt_lines, ws_lines, fromfile=f"a/{rel_path}", tofile=f"b/{rel_path}", lineterm=""))
+            if diff:
+                files_changed.append({"file": rel_path, "status": "modified"})
+                diff_lines.extend(diff)
+
+    # Check for new files in workspace that aren't in checkpoint
+    # (skip for performance — diff is primarily for seeing what the checkpoint captures)
+
+    return {
+        "checkpoint": checkpoint,
+        "workspace": resolved,
+        "diff": "\n".join(diff_lines) if diff_lines else "",
+        "files_changed": files_changed,
+        "total_changes": len(files_changed),
+    }
+
+
+def restore_checkpoint(workspace: str, checkpoint: str) -> dict[str, Any]:
+    """Restore a checkpoint by copying files back to the workspace.
+
+    Only restores files that exist in the checkpoint.  Does NOT delete
+    files that were added after the checkpoint was created.
+
+    Returns a dict with:
+        ok: True
+        files_restored: list of restored file paths
+    """
+    resolved = _resolve_workspace(workspace)
+    checkpoint = _validate_checkpoint_id(checkpoint)
+    ws_hash = _workspace_hash(resolved)
+    ckpt_dir = _checkpoint_root() / ws_hash / checkpoint
+
+    if not ckpt_dir.is_dir():
+        raise ValueError(f"Checkpoint not found: {checkpoint}")
+
+    git = _find_git()
+
+    # Get list of files in the checkpoint
+    ls_result = subprocess.run(
+        [git, "-C", str(ckpt_dir), "ls-files"],
+        capture_output=True, text=True, timeout=10,
+    )
+    if ls_result.returncode != 0:
+        raise ValueError("Failed to list checkpoint files")
+
+    ckpt_files = [f for f in ls_result.stdout.strip().split("\n") if f]
+    restored = []
+    errors = []
+
+    for rel_path in ckpt_files:
+        ckpt_file = ckpt_dir / rel_path
+        ws_file = Path(resolved) / rel_path
+
+        if not ckpt_file.is_file():
+            continue
+
+        try:
+            ws_file.parent.mkdir(parents=True, exist_ok=True)
+            shutil.copy2(str(ckpt_file), str(ws_file))
+            restored.append(rel_path)
+        except OSError as e:
+            errors.append({"file": rel_path, "error": str(e)})
+            logger.warning("Failed to restore %s: %s", rel_path, e)
+
+    return {
+        "ok": True,
+        "checkpoint": checkpoint,
+        "workspace": resolved,
+        "files_restored": restored,
+        "files_restored_count": len(restored),
+        "errors": errors,
+    }
diff --git a/api/routes.py b/api/routes.py
index 4237f37d..5440a108 100644
--- a/api/routes.py
+++ b/api/routes.py
@@ -215,6 +215,7 @@ from api.config import (
     load_settings,
     save_settings,
     set_hermes_default_model,
+    model_with_provider_context,
     get_reasoning_status,
     set_reasoning_display,
     set_reasoning_effort,
@@ -397,20 +398,78 @@ def _model_matches_active_provider_family(
     return False
 
 
-def _resolve_compatible_session_model(model_id: str | None) -> tuple[str, bool]:
-    """Return (effective_model, was_normalized) for persisted session models.
+def _catalog_model_id_matches(candidate: str, model: str) -> bool:
+    candidate = str(candidate or "").strip()
+    if candidate.startswith("@") and ":" in candidate:
+        candidate = candidate.rsplit(":", 1)[1]
+    if "/" in candidate:
+        candidate = candidate.split("/", 1)[1]
+    return candidate.replace("-", ".").lower() == model.replace("-", ".").lower()
+
+
+def _clean_session_model_provider(value: str | None) -> str | None:
+    provider = str(value or "").strip().lower()
+    if not provider or provider == "default":
+        return None
+    if provider.startswith("@"):
+        provider = provider[1:]
+    return provider or None
+
+
+def _split_provider_qualified_model(model: str) -> tuple[str, str | None]:
+    model = str(model or "").strip()
+    if model.startswith("@") and ":" in model:
+        provider_hint, bare_model = model[1:].rsplit(":", 1)
+        provider = _clean_session_model_provider(provider_hint)
+        bare = bare_model.strip()
+        if provider and bare:
+            return bare, provider
+    return model, None
+
+
+def _should_attach_codex_provider_context(model: str, raw_active_provider: str, catalog: dict) -> bool:
+    """Return True when a bare Codex model needs separate provider context.
+
+    OpenAI, OpenAI Codex, Copilot, and OpenRouter can all expose GPT-looking
+    bare names. If a session stores only ``gpt-...`` while Codex is active, a
+    later provider-list/default-model round trip can lose the user's Codex
+    choice. Store the provider separately instead of converting the persisted
+    model to ``@openai-codex:model``.
+    """
+    if raw_active_provider != "openai-codex":
+        return False
+    if not model.lower().startswith("gpt"):
+        return False
+    for group in catalog.get("groups") or []:
+        if str(group.get("provider_id") or "").strip().lower() != "openai-codex":
+            continue
+        return any(
+            _catalog_model_id_matches(entry.get("id"), model)
+            for entry in group.get("models", [])
+            if isinstance(entry, dict)
+        )
+    return False
+
+
+def _resolve_compatible_session_model_state(
+    model_id: str | None,
+    model_provider: str | None = None,
+) -> tuple[str, str | None, bool]:
+    """Return (effective_model, effective_provider, model_was_normalized).
 
     Sessions can outlive provider changes. When an older session still points at
     a different provider namespace (for example `gemini/...` after switching the
     agent to OpenAI Codex), reusing that stale model causes chat startup to hit
-    the wrong backend and fail. Normalize only obvious cross-provider mismatches;
-    preserve bare model IDs and OpenRouter/custom setups.
+    the wrong backend and fail. Normalize only obvious cross-provider mismatches.
+    When a model has an explicit provider context, keep the model string itself
+    in its picker/API shape and carry the provider as separate state.
     """
     catalog = get_available_models()
     default_model = str(catalog.get("default_model") or DEFAULT_MODEL or "").strip()
     model = str(model_id or "").strip()
+    requested_provider = _clean_session_model_provider(model_provider)
     if not model:
-        return default_model, bool(default_model)
+        return default_model, requested_provider, bool(default_model)
 
     active_provider = _normalize_provider_id(catalog.get("active_provider"))
     # Also keep the raw active_provider slug for cross-provider detection with
@@ -420,15 +479,19 @@ def _resolve_compatible_session_model(model_id: str | None) -> tuple[str, bool]:
     # is stale relative to this unknown active provider. (#1023)
     raw_active_provider = str(catalog.get("active_provider") or "").strip().lower()
     if not active_provider and not raw_active_provider:
-        return model, False
+        bare_model, explicit_provider = _split_provider_qualified_model(model)
+        return model, explicit_provider or requested_provider, False
+
+    bare_for_context, explicit_provider = _split_provider_qualified_model(model)
+    if requested_provider and not explicit_provider:
+        return model, requested_provider, False
 
     if model.startswith("@") and ":" in model:
-        provider_hint, bare_model = model[1:].split(":", 1)
-        provider_raw = provider_hint.strip().lower()
+        provider_raw = explicit_provider or ""
         provider_normalized = _normalize_provider_id(provider_raw)
-        bare_model = bare_model.strip()
+        bare_model = bare_for_context.strip()
         if not provider_raw or not bare_model:
-            return model, False
+            return model, requested_provider, False
 
         raw_provider_ids, normalized_provider_ids = _catalog_provider_id_sets(catalog)
         hint_matches_active = (
@@ -444,7 +507,7 @@ def _resolve_compatible_session_model(model_id: str | None) -> tuple[str, bool]:
             # here would collapse duplicate model IDs from different providers back to the
             # bare ID, causing the first matching provider to win on the next UI render
             # and the wrong provider to be used for the agent run. (#1253)
-            return model, False
+            return model, provider_raw, False
 
         if _catalog_has_provider(
             provider_raw,
@@ -452,13 +515,23 @@ def _resolve_compatible_session_model(model_id: str | None) -> tuple[str, bool]:
             raw_provider_ids,
             normalized_provider_ids,
         ):
-            return model, False
+            return model, provider_raw, False
 
         if _model_matches_active_provider_family(bare_model, active_provider):
-            return bare_model, True
+            provider_context = (
+                raw_active_provider
+                if _should_attach_codex_provider_context(bare_model, raw_active_provider, catalog)
+                else None
+            )
+            return bare_model, provider_context, True
         if default_model:
-            return default_model, True
-        return model, False
+            provider_context = (
+                raw_active_provider
+                if _should_attach_codex_provider_context(default_model, raw_active_provider, catalog)
+                else None
+            )
+            return default_model, provider_context, True
+        return model, provider_raw, False
 
     slash = model.find("/")
     if slash < 0:
@@ -467,9 +540,19 @@ def _resolve_compatible_session_model(model_id: str | None) -> tuple[str, bool]:
             if model_lower.startswith(bare_prefix):
                 model_provider = _normalize_provider_id(bare_prefix)
                 if model_provider and model_provider != active_provider and default_model:
-                    return default_model, True
-                return model, False
-        return model, False
+                    provider_context = (
+                        raw_active_provider
+                        if _should_attach_codex_provider_context(default_model, raw_active_provider, catalog)
+                        else None
+                    )
+                    return default_model, provider_context, True
+                provider_context = (
+                    raw_active_provider
+                    if _should_attach_codex_provider_context(model, raw_active_provider, catalog)
+                    else requested_provider
+                )
+                return model, provider_context, False
+        return model, requested_provider, False
 
     model_provider = _normalize_provider_id(model[:slash])
 
@@ -481,7 +564,7 @@ def _resolve_compatible_session_model(model_id: str | None) -> tuple[str, bool]:
     if active_provider in {"custom", "openrouter"}:
         # These namespaces are always routable as-is — preserve them.
         if model_provider in {"", "custom", "openrouter"}:
-            return model, False
+            return model, requested_provider, False
         # Check if any catalog group can actually route this model's prefix.
         groups = catalog.get("groups") or []
         routable_provider_ids = {
@@ -492,11 +575,11 @@ def _resolve_compatible_session_model(model_id: str | None) -> tuple[str, bool]:
             (g.get("provider_id") or "") == "openrouter" for g in groups
         )
         if model_provider in routable_provider_ids or has_openrouter_group:
-            return model, False
+            return model, requested_provider, False
         # Model prefix is not routable — stale cross-provider reference, clear it.
         if default_model:
-            return default_model, True
-        return model, False
+            return default_model, requested_provider, True
+        return model, requested_provider, False
 
     # Skip normalization for models on custom/openrouter namespaces — these are
     # user-controlled and should never be silently replaced.
@@ -506,18 +589,35 @@ def _resolve_compatible_session_model(model_id: str | None) -> tuple[str, bool]:
     # active provider name, the session model is stale. (#1023)
     _active_for_compare = active_provider or raw_active_provider
     if model_provider and model_provider not in {"", "custom", "openrouter"} and model_provider != _active_for_compare and default_model:
-        return default_model, True
-    return model, False
+        return default_model, requested_provider, True
+    return model, requested_provider, False
+
+
+def _resolve_compatible_session_model(model_id: str | None) -> tuple[str, bool]:
+    """Return (effective_model, model_was_normalized) for legacy callers."""
+    effective_model, _provider, changed = _resolve_compatible_session_model_state(model_id)
+    return effective_model, changed
 
 
 def _normalize_session_model_in_place(session) -> str:
     original_model = getattr(session, "model", None) or ""
-    effective_model, changed = _resolve_compatible_session_model(original_model or None)
+    original_provider = _clean_session_model_provider(
+        getattr(session, "model_provider", None)
+    )
+    effective_model, effective_provider, changed = _resolve_compatible_session_model_state(
+        original_model or None,
+        original_provider,
+    )
+    provider_changed = effective_provider != original_provider
     # Only persist the correction if the session had an explicit model that needed changing.
     # Sessions with no model stored (empty/None) get the effective default returned without
     # a disk write — no need to rebuild the index for a fill-in-blank operation.
-    if changed and effective_model and original_model and original_model != effective_model:
-        session.model = effective_model
+    if original_model and effective_model and (
+        (changed and original_model != effective_model) or provider_changed
+    ):
+        if changed and original_model != effective_model:
+            session.model = effective_model
+        session.model_provider = effective_provider
         session.save(touch_updated_at=False)
     return effective_model
 
@@ -530,10 +630,46 @@ def _resolve_effective_session_model_for_display(session) -> str:
     effective model for the response payload only and leave disk state alone.
     """
     original_model = getattr(session, "model", None) or ""
-    effective_model, _changed = _resolve_compatible_session_model(original_model or None)
+    effective_model, _provider, _changed = _resolve_compatible_session_model_state(
+        original_model or None,
+        getattr(session, "model_provider", None),
+    )
     return effective_model or original_model
 
 
+def _resolve_effective_session_model_provider_for_display(session) -> str | None:
+    original_model = getattr(session, "model", None) or ""
+    _model, provider, _changed = _resolve_compatible_session_model_state(
+        original_model or None,
+        getattr(session, "model_provider", None),
+    )
+    return provider
+
+
+def _session_model_state_from_request(
+    model: str | None,
+    requested_provider: str | None,
+    current_provider: str | None = None,
+) -> tuple[str | None, str | None]:
+    model_value = str(model).strip() if model is not None else None
+    provider = (
+        _clean_session_model_provider(requested_provider)
+        if requested_provider is not None
+        else None
+    )
+    if model_value:
+        _bare, explicit_provider = _split_provider_qualified_model(model_value)
+        if explicit_provider:
+            provider = explicit_provider
+        elif requested_provider is None:
+            provider = _clean_session_model_provider(current_provider)
+        model_value, provider, _changed = _resolve_compatible_session_model_state(
+            model_value,
+            provider,
+        )
+    return model_value, provider
+
+
 from api.models import (
     Session,
     get_session,
@@ -842,6 +978,102 @@ button:hover{background:rgba(124,185,255,.25)}
 <script src="/static/login.js"></script>
 </body></html>"""
 
+# ── Insights endpoint ──────────────────────────────────────────────────────────
+
+def _handle_insights(handler, parsed) -> bool:
+    """Return usage analytics from local WebUI session data."""
+    import collections
+    import time as _time
+
+    query = parse_qs(parsed.query)
+    try:
+        days = min(max(int(query.get("days", ["30"])[0]), 1), 365)
+    except (ValueError, TypeError):
+        days = 30
+
+    now = _time.time()
+    cutoff = now - (days * 86400)
+
+    # Walk session index (fast, no full JSON parse)
+    sessions_data = []
+    idx_path = SESSION_DIR / "_index.json"
+    if idx_path.exists():
+        try:
+            idx = json.loads(idx_path.read_text(encoding="utf-8"))
+        except Exception:
+            idx = []
+    else:
+        idx = []
+
+    for entry in idx:
+        created = entry.get("created_at", 0) or 0
+        updated = entry.get("updated_at", 0) or 0
+        # Session is relevant if it was created or updated within the window
+        if max(created, updated) < cutoff:
+            continue
+        sessions_data.append(entry)
+
+    # Aggregate
+    total_sessions = len(sessions_data)
+    total_messages = 0
+    total_input_tokens = 0
+    total_output_tokens = 0
+    total_cost = 0.0
+    model_counts = collections.Counter()
+    # Activity by day of week (0=Mon .. 6=Sun)
+    dow_activity = collections.Counter()
+    # Activity by hour of day (0-23)
+    hod_activity = collections.Counter()
+
+    for s in sessions_data:
+        total_messages += max(s.get("message_count", 0) or 0, 0)
+        total_input_tokens += max(s.get("input_tokens", 0) or 0, 0)
+        total_output_tokens += max(s.get("output_tokens", 0) or 0, 0)
+        cost = s.get("estimated_cost")
+        if cost is not None:
+            try:
+                total_cost += float(cost)
+            except (ValueError, TypeError):
+                pass
+        model = s.get("model") or "unknown"
+        if model:
+            model_counts[model] += 1
+        # Activity patterns
+        ts = s.get("updated_at", s.get("created_at", 0)) or 0
+        if ts:
+            try:
+                dt = _time.localtime(ts)
+                dow_activity[dt.tm_wday] += 1
+                hod_activity[dt.tm_hour] += 1
+            except Exception:
+                pass
+
+    # Build model breakdown
+    models_breakdown = []
+    for model, count in model_counts.most_common():
+        models_breakdown.append({"model": model, "sessions": count})
+
+    # Day-of-week labels
+    dow_labels = ["Mon", "Tue", "Wed", "Thu", "Fri", "Sat", "Sun"]
+    dow_data = [{"day": dow_labels[i], "sessions": dow_activity.get(i, 0)} for i in range(7)]
+
+    # Hour-of-day data
+    hod_data = [{"hour": h, "sessions": hod_activity.get(h, 0)} for h in range(24)]
+
+    return j(handler, {
+        "period_days": days,
+        "total_sessions": total_sessions,
+        "total_messages": total_messages,
+        "total_input_tokens": total_input_tokens,
+        "total_output_tokens": total_output_tokens,
+        "total_tokens": total_input_tokens + total_output_tokens,
+        "total_cost": round(total_cost, 6),
+        "models": models_breakdown,
+        "activity_by_day": dow_data,
+        "activity_by_hour": hod_data,
+    })
+
+
 # ── GET routes ────────────────────────────────────────────────────────────────
 
 
@@ -944,6 +1176,10 @@ def handle_get(handler, parsed) -> bool:
             handler.end_headers()
         return True
 
+    # ── Insights ──
+    if parsed.path == "/api/insights":
+        return _handle_insights(handler, parsed)
+
     if parsed.path == "/health":
         with STREAMS_LOCK:
             n_streams = len(STREAMS)
@@ -1033,6 +1269,11 @@ def handle_get(handler, parsed) -> bool:
                 if resolve_model
                 else None
             )
+            effective_provider = (
+                _resolve_effective_session_model_provider_for_display(s)
+                if resolve_model
+                else None
+            )
             _t3 = _time.monotonic()
             _all_msgs = s.messages if load_messages else []
             if load_messages:
@@ -1080,6 +1321,8 @@ def handle_get(handler, parsed) -> bool:
             _t4 = _time.monotonic()
             if effective_model:
                 raw["model"] = effective_model
+            if effective_provider:
+                raw["model_provider"] = effective_provider
             redact = redact_session_data(raw)
             _t5 = _time.monotonic()
             resp = j(handler, {"session": redact})
@@ -1428,10 +1671,40 @@ def handle_get(handler, parsed) -> bool:
     if parsed.path == "/api/mcp/servers":
         return _handle_mcp_servers_list(handler)
 
+    # ── Checkpoints / Rollback (GET) ──
+    if parsed.path == "/api/rollback/list":
+        qs = parse_qs(parsed.query)
+        workspace = qs.get("workspace", [""])[0]
+        if not workspace:
+            return bad(handler, "workspace query parameter is required")
+        try:
+            from api.rollback import list_checkpoints
+            return j(handler, list_checkpoints(workspace))
+        except ValueError as e:
+            return bad(handler, str(e))
+        except Exception as e:
+            logger.exception("rollback/list failed")
+            return bad(handler, str(e), status=500)
+
+    if parsed.path == "/api/rollback/diff":
+        qs = parse_qs(parsed.query)
+        workspace = qs.get("workspace", [""])[0]
+        checkpoint = qs.get("checkpoint", [""])[0]
+        if not workspace or not checkpoint:
+            return bad(handler, "workspace and checkpoint query parameters are required")
+        try:
+            from api.rollback import get_checkpoint_diff
+            return j(handler, get_checkpoint_diff(workspace, checkpoint))
+        except ValueError as e:
+            return bad(handler, str(e))
+        except Exception as e:
+            logger.exception("rollback/diff failed")
+            return bad(handler, str(e), status=500)
+
     return False  # 404
 
 
-# ── POST routes ───────────────────────────────────────────────────────────────
+# ── GET route helpers
 
 
 def handle_post(handler, parsed) -> bool:
@@ -1455,9 +1728,18 @@ def handle_post(handler, parsed) -> bool:
             workspace = str(resolve_trusted_workspace(body.get("workspace"))) if body.get("workspace") else None
         except ValueError as e:
             return bad(handler, str(e))
+        model, model_provider = _session_model_state_from_request(
+            body.get("model"),
+            body.get("model_provider"),
+        )
         # Use the profile sent by the client tab (if any) so that two tabs on
         # different profiles never clobber each other via the process-level global.
-        s = new_session(workspace=workspace, model=body.get("model"), profile=body.get("profile") or None)
+        s = new_session(
+            workspace=workspace,
+            model=model,
+            model_provider=model_provider,
+            profile=body.get("profile") or None,
+        )
         return j(handler, {"session": s.compact() | {"messages": s.messages}})
 
     if parsed.path == "/api/default-model":
@@ -1610,7 +1892,15 @@ def handle_post(handler, parsed) -> bool:
             return bad(handler, str(e))
         with _get_session_agent_lock(body["session_id"]):
             s.workspace = new_ws
-            s.model = body.get("model", s.model)
+            if "model" in body or "model_provider" in body:
+                model, provider = _session_model_state_from_request(
+                    body.get("model", s.model),
+                    body.get("model_provider") if "model_provider" in body else None,
+                    getattr(s, "model_provider", None),
+                )
+                if model is not None:
+                    s.model = model
+                s.model_provider = provider
             s.save()
         if str(old_ws or "") != str(new_ws or ""):
             try:
@@ -2289,6 +2579,23 @@ def handle_post(handler, parsed) -> bool:
         handler.wfile.write(json.dumps({"ok": True}).encode())
         return True
 
+    # ── Checkpoints / Rollback (POST) ──
+    if parsed.path == "/api/rollback/restore":
+        if not body:
+            return bad(handler, "request body is required")
+        workspace = body.get("workspace", "")
+        checkpoint = body.get("checkpoint", "")
+        if not workspace or not checkpoint:
+            return bad(handler, "workspace and checkpoint are required")
+        try:
+            from api.rollback import restore_checkpoint
+            return j(handler, restore_checkpoint(workspace, checkpoint))
+        except ValueError as e:
+            return bad(handler, str(e))
+        except Exception as e:
+            logger.exception("rollback/restore failed")
+            return bad(handler, str(e), status=500)
+
     return False  # 404
 
 # ── GET route helpers ─────────────────────────────────────────────────────────
@@ -3473,7 +3780,13 @@ def _handle_btw(handler, body):
         s.active_stream_id = None
     # Create ephemeral hidden session inheriting context
     from api.models import new_session as _new_session
-    ephemeral = _new_session(workspace=s.workspace, model=s.model, profile=getattr(s, 'profile', None))
+    model_provider = getattr(s, 'model_provider', None)
+    ephemeral = _new_session(
+        workspace=s.workspace,
+        model=s.model,
+        model_provider=model_provider,
+        profile=getattr(s, 'profile', None),
+    )
     # Copy conversation history for context (agent reads from messages)
     ephemeral.messages = list(s.messages or [])
     ephemeral.title = f"btw: {question[:60]}"
@@ -3489,7 +3802,7 @@ def _handle_btw(handler, body):
     thr = threading.Thread(
         target=_run_agent_streaming,
         args=(ephemeral.session_id, question, s.model, s.workspace, stream_id, None),
-        kwargs={"ephemeral": True},
+        kwargs={"ephemeral": True, "model_provider": model_provider},
         daemon=True,
     )
     thr.start()
@@ -3515,7 +3828,13 @@ def _handle_background(handler, body):
     if not prompt:
         return bad(handler, "prompt is required")
     from api.models import new_session as _new_session
-    bg = _new_session(workspace=s.workspace, model=s.model, profile=getattr(s, 'profile', None))
+    model_provider = getattr(s, 'model_provider', None)
+    bg = _new_session(
+        workspace=s.workspace,
+        model=s.model,
+        model_provider=model_provider,
+        profile=getattr(s, 'profile', None),
+    )
     bg.title = f"bg: {prompt[:60]}"
     bg.save()
     stream_id = uuid.uuid4().hex
@@ -3537,7 +3856,15 @@ def _handle_background(handler, body):
         `get_results()` would see a forever-`running` task and return nothing.
         """
         try:
-            _run_agent_streaming(bg_sid, prompt, s.model, s.workspace, stream_id, None)
+            _run_agent_streaming(
+                bg_sid,
+                prompt,
+                s.model,
+                s.workspace,
+                stream_id,
+                None,
+                model_provider=model_provider,
+            )
             # Reload the bg session from disk and extract the final assistant reply.
             try:
                 from api.models import Session as _Session
@@ -3591,7 +3918,15 @@ def _handle_chat_start(handler, body):
     except ValueError as e:
         return bad(handler, str(e))
     requested_model = body.get("model") or s.model
-    model, normalized_model = _resolve_compatible_session_model(requested_model)
+    requested_provider = (
+        body.get("model_provider")
+        if "model_provider" in body
+        else getattr(s, "model_provider", None)
+    )
+    model, model_provider, normalized_model = _resolve_compatible_session_model_state(
+        requested_model,
+        requested_provider,
+    )
     # Prevent duplicate runs in the same session while a stream is still active.
     # This commonly happens after page refresh/reconnect races and can produce
     # duplicated clarify cards for what appears to be a single user request.
@@ -3614,6 +3949,7 @@ def _handle_chat_start(handler, body):
     with _get_session_agent_lock(s.session_id):
         s.workspace = workspace
         s.model = model
+        s.model_provider = model_provider
         s.active_stream_id = stream_id
         s.pending_user_message = msg
         s.pending_attachments = attachments
@@ -3626,12 +3962,15 @@ def _handle_chat_start(handler, body):
     thr = threading.Thread(
         target=_run_agent_streaming,
         args=(s.session_id, msg, model, workspace, stream_id, attachments),
+        kwargs={"model_provider": model_provider},
         daemon=True,
     )
     thr.start()
     response = {"stream_id": stream_id, "session_id": s.session_id}
     if normalized_model:
         response["effective_model"] = model
+    if model_provider:
+        response["effective_model_provider"] = model_provider
     return j(handler, response)
 
 
@@ -3677,7 +4016,12 @@ def _handle_chat_sync(handler, body):
         return bad(handler, str(e))
     with _get_session_agent_lock(s.session_id):
         s.workspace = workspace
-        s.model = body.get("model") or s.model
+        model, model_provider = _resolve_compatible_session_model_state(
+            body.get("model") or s.model,
+            body.get("model_provider") if "model_provider" in body else getattr(s, "model_provider", None),
+        )[:2]
+        s.model = model
+        s.model_provider = model_provider
     from api.streaming import _ENV_LOCK
 
     with _ENV_LOCK:
@@ -3693,7 +4037,9 @@ def _handle_chat_sync(handler, body):
         with CHAT_LOCK:
             from api.config import resolve_model_provider
 
-            _model, _provider, _base_url = resolve_model_provider(s.model)
+            _model, _provider, _base_url = resolve_model_provider(
+                model_with_provider_context(s.model, getattr(s, "model_provider", None))
+            )
             # Resolve API key via Hermes runtime provider (matches gateway behaviour)
             _api_key = None
             try:
@@ -4356,7 +4702,9 @@ def _handle_session_compress(handler, body):
         import hermes_cli.runtime_provider as _runtime_provider
         import run_agent as _run_agent
 
-        resolved_model, resolved_provider, resolved_base_url = _cfg.resolve_model_provider(s.model)
+        resolved_model, resolved_provider, resolved_base_url = _cfg.resolve_model_provider(
+            _cfg.model_with_provider_context(s.model, getattr(s, "model_provider", None))
+        )
 
         resolved_api_key = None
         try:
diff --git a/api/streaming.py b/api/streaming.py
index 4ed12b83..1abbea64 100644
--- a/api/streaming.py
+++ b/api/streaming.py
@@ -26,6 +26,7 @@ from api.config import (
     _get_session_agent_lock, _set_thread_env, _clear_thread_env,
     SESSION_AGENT_LOCKS, SESSION_AGENT_LOCKS_LOCK,
     resolve_model_provider,
+    model_with_provider_context,
 )
 from api.helpers import redact_session_data
 from api.metering import meter
@@ -1342,7 +1343,17 @@ def _last_resort_sync_from_core(session, stream_id, agent_lock):
         )
 
 
-def _run_agent_streaming(session_id, msg_text, model, workspace, stream_id, attachments=None, *, ephemeral=False):
+def _run_agent_streaming(
+    session_id,
+    msg_text,
+    model,
+    workspace,
+    stream_id,
+    attachments=None,
+    *,
+    ephemeral=False,
+    model_provider=None,
+):
     """Run agent in background thread, writing SSE events to STREAMS[stream_id].
 
     When ephemeral=True, session mutations are skipped — used by /btw to get
@@ -1418,6 +1429,12 @@ def _run_agent_streaming(session_id, msg_text, model, workspace, stream_id, atta
         s = get_session(session_id)
         s.workspace = str(Path(workspace).expanduser().resolve())
         s.model = model
+        provider_context = (
+            str(model_provider).strip().lower()
+            if model_provider is not None
+            else getattr(s, "model_provider", None)
+        )
+        s.model_provider = provider_context or None
 
         _agent_lock = _get_session_agent_lock(session_id)
         # TD1: set thread-local env context so concurrent sessions don't clobber globals
@@ -1701,7 +1718,9 @@ def _run_agent_streaming(session_id, msg_text, model, workspace, stream_id, atta
                 _session_db = SessionDB()
             except Exception as _db_err:
                 print(f"[webui] WARNING: SessionDB init failed — session_search will be unavailable: {_db_err}", flush=True)
-            resolved_model, resolved_provider, resolved_base_url = resolve_model_provider(model)
+            resolved_model, resolved_provider, resolved_base_url = resolve_model_provider(
+                model_with_provider_context(model, provider_context)
+            )
 
             # Resolve API key via Hermes runtime provider (matches gateway behaviour).
             # Pass the resolved provider so non-default providers get their own credentials.
diff --git a/static/boot.js b/static/boot.js
index a15b1792..e4134258 100644
--- a/static/boot.js
+++ b/static/boot.js
@@ -199,6 +199,12 @@ $('btnSend').onclick=()=>{
     _stopMic();
     return;
   }
+  // Turn-based voice mode: let the voice mode system handle the send flow
+  if(typeof window._voiceModeActive==='function'&&window._voiceModeActive()){
+    // Immediately send whatever is in the textarea
+    if(typeof window._voiceModeImmediateSend==='function') window._voiceModeImmediateSend();
+    return;
+  }
   send();
 };
 $('btnAttach').onclick=()=>$('fileInput').click();
@@ -403,6 +409,284 @@ $('btnAttach').onclick=()=>$('fileInput').click();
 })();
 window._micActive=window._micActive||false;
 window._micPendingSend=window._micPendingSend||false;
+
+// ── Turn-based voice mode (#1333) ────────────────────────────────────────
+// Chained flow: listen → send → (agent processes) → TTS response → listen again
+(function(){
+  const SpeechRecognition=window.SpeechRecognition||window.webkitSpeechRecognition;
+  const hasSTT=!(!SpeechRecognition);
+  const hasTTS=!!('speechSynthesis' in window);
+
+  // Need both STT and TTS for turn-based voice mode
+  if(!hasSTT||!hasTTS) return;
+
+  const modeBtn=$('btnVoiceMode');
+  const bar=$('voiceModeBar');
+  const indicator=$('voiceModeIndicator');
+  const label=$('voiceModeLabel');
+  const micBtn=$('btnMic');
+  const ta=$('msg');
+
+  if(!modeBtn||!bar||!indicator||!label) return;
+
+  // Show the voice mode button — browser supports both STT and TTS
+  modeBtn.style.display='';
+
+  let _voiceModeActive=false;
+  let _voiceModeState='idle'; // idle | listening | thinking | speaking
+  let _recognition=null;
+  let _silenceTimer=null;
+  // Capture the session id at thinking-time so the TTS callback won't read
+  // a different session's last assistant reply if the user navigated away
+  // between send and stream completion. (Opus pre-release advisor.)
+  let _voiceModeThinkingSid=null;
+  const SILENCE_MS=1800; // auto-send after 1.8s silence
+
+  function _setState(state){
+    _voiceModeState=state;
+    indicator.className='voice-mode-indicator '+state;
+    label.textContent=state==='listening'?t('voice_listening')
+      :state==='speaking'?t('voice_speaking')
+      :state==='thinking'?t('voice_thinking')
+      :'';
+    bar.style.display=_voiceModeActive?(state==='idle'?'none':''):'none';
+  }
+
+  function _startListening(){
+    if(!_voiceModeActive) return;
+    _setState('listening');
+
+    _recognition=new SpeechRecognition();
+    _recognition.continuous=false;
+    _recognition.interimResults=true;
+    _recognition.lang=(typeof _locale!=='undefined'&&_locale._speech)||'en-US';
+
+    let _finalText='';
+
+    _recognition.onstart=()=>{ _finalText=''; };
+
+    _recognition.onresult=(event)=>{
+      // Reset silence timer on any result
+      clearTimeout(_silenceTimer);
+      let interim='';
+      let final=_finalText;
+      for(let i=event.resultIndex;i<event.results.length;i++){
+        const txt=event.results[i][0].transcript;
+        if(event.results[i].isFinal){ final+=txt; _finalText=final; }
+        else{ interim+=txt; }
+      }
+      ta.value=final||interim;
+      autoResize();
+
+      // Auto-send on silence after final result
+      if(_finalText){
+        _silenceTimer=setTimeout(()=>{
+          _voiceModeSend();
+        },SILENCE_MS);
+      }
+    };
+
+    _recognition.onend=()=>{
+      clearTimeout(_silenceTimer);
+      // If we have text and haven't sent yet, send it
+      if(_finalText&&_voiceModeActive&&_voiceModeState==='listening'){
+        _voiceModeSend();
+      } else if(_voiceModeActive&&_voiceModeState==='listening'){
+        // No speech detected — restart listening
+        setTimeout(()=>{ if(_voiceModeActive) _startListening(); },500);
+      }
+    };
+
+    _recognition.onerror=(event)=>{
+      clearTimeout(_silenceTimer);
+      if(event.error==='no-speech'||event.error==='aborted'){
+        // Restart if still active
+        if(_voiceModeActive){
+          setTimeout(()=>{ if(_voiceModeActive) _startListening(); },800);
+        }
+        return;
+      }
+      if(event.error==='not-allowed'||event.error==='service-not-allowed'||event.error==='audio-capture'){
+        _deactivate();
+        showToast(t('mic_denied'));
+        return;
+      }
+      // Other errors — try to restart
+      if(_voiceModeActive){
+        setTimeout(()=>{ if(_voiceModeActive) _startListening(); },1500);
+      }
+    };
+
+    try{ _recognition.start(); }catch(e){
+      // Already started or other error — retry shortly
+      setTimeout(()=>{ if(_voiceModeActive) _startListening(); },1000);
+    }
+  }
+
+  function _voiceModeSend(){
+    if(!_voiceModeActive) return;
+    const text=(ta.value||'').trim();
+    if(!text){
+      ta.value='';
+      setTimeout(()=>{ if(_voiceModeActive) _startListening(); },300);
+      return;
+    }
+    _setState('thinking');
+    // Pin the active session id so the TTS callback won't speak a different
+    // session's reply if the user navigates away mid-stream.
+    _voiceModeThinkingSid=(typeof S!=='undefined'&&S.session)?S.session.session_id:null;
+    try{ if(_recognition) _recognition.abort(); }catch(_){}
+    _recognition=null;
+    // send() is global from boot.js
+    if(typeof send==='function') send();
+  }
+
+  function _speakResponse(){
+    if(!_voiceModeActive) return;
+    // Bail out if the user navigated to a different session between send and
+    // stream completion. The patched autoReadLastAssistant fires globally;
+    // without this guard it would TTS-read the wrong session's last assistant
+    // message. Drop back to listening on the new session instead.
+    const currentSid=(typeof S!=='undefined'&&S.session)?S.session.session_id:null;
+    if(_voiceModeThinkingSid && currentSid && currentSid!==_voiceModeThinkingSid){
+      _voiceModeThinkingSid=null;
+      _startListening();
+      return;
+    }
+    _voiceModeThinkingSid=null;
+    _setState('speaking');
+
+    // Find last assistant message
+    const rows=document.querySelectorAll('.msg-row[data-role="assistant"], .assistant-segment[data-raw-text]');
+    if(!rows.length){ _startListening(); return; }
+    const last=rows[rows.length-1];
+    const rawText=last.dataset.rawText||'';
+    if(!rawText.trim()){ _startListening(); return; }
+
+    // Strip for TTS (reuse existing helper if available)
+    let clean=rawText;
+    if(typeof _stripForTTS==='function') clean=_stripForTTS(rawText);
+    else{
+      // Basic strip: remove code blocks, images, links
+      clean=clean.replace(/```[\s\S]*?```/g,' code block ')
+        .replace(/`([^`]*)`/g,'$1')
+        .replace(/!\[([^\]]*)\]\([^)]*\)/g,'$1')
+        .replace(/\[([^\]]*)\]\([^)]*\)/g,'$1')
+        .replace(/#{1,6}\s/g,'')
+        .replace(/[*_~]+/g,'')
+        .replace(/\n{2,}/g,'. ')
+        .replace(/\n/g,' ')
+        .trim();
+    }
+    if(!clean){ _startListening(); return; }
+
+    const utter=new SpeechSynthesisUtterance(clean);
+
+    // Apply saved voice preferences
+    const savedVoice=localStorage.getItem('hermes-tts-voice');
+    const voices=speechSynthesis.getVoices();
+    if(savedVoice&&voices.length){
+      const match=voices.find(v=>v.name===savedVoice);
+      if(match) utter.voice=match;
+    }
+    const savedRate=parseFloat(localStorage.getItem('hermes-tts-rate'));
+    if(!isNaN(savedRate)) utter.rate=Math.min(2,Math.max(0.5,savedRate));
+    const savedPitch=parseFloat(localStorage.getItem('hermes-tts-pitch'));
+    if(!isNaN(savedPitch)) utter.pitch=Math.min(2,Math.max(0,savedPitch));
+
+    utter.onend=()=>{
+      // After speaking, go back to listening
+      if(_voiceModeActive) setTimeout(()=>_startListening(),500);
+    };
+    utter.onerror=()=>{
+      if(_voiceModeActive) setTimeout(()=>_startListening(),1000);
+    };
+
+    speechSynthesis.speak(utter);
+  }
+
+  // Hook into response completion — observe when the agent finishes
+  // We patch setComposerStatus to detect when a response completes
+  const _origSetComposerStatus=(typeof setComposerStatus==='function')?setComposerStatus.bind(window):null;
+
+  window._voiceModeOnResponseComplete=function(){
+    if(_voiceModeActive&&_voiceModeState==='thinking'){
+      // Small delay to let DOM render the final message
+      setTimeout(()=>{
+        if(_voiceModeActive&&_voiceModeState==='thinking'){
+          _speakResponse();
+        }
+      },400);
+    }
+  };
+
+  // Observe S.busy changes to detect response completion
+  // The existing code calls setBusy(false) when response completes
+  const _origSetBusy=(typeof setBusy==='function')?setBusy.bind(window):null;
+  if(_origSetBusy){
+    // We use a MutationObserver-style approach via polling S.busy
+    // Actually, we'll use a simpler approach: hook into the message stream completion
+  }
+
+  // Most reliable hook: use the existing autoReadLastAssistant call site.
+  // We override autoReadLastAssistant so that if voice mode is active, we use our
+  // own speak-and-resume flow instead of the default auto-read.
+  const _origAutoRead=(typeof autoReadLastAssistant==='function')?autoReadLastAssistant:null;
+  window.autoReadLastAssistant=function(){
+    if(_voiceModeActive&&_voiceModeState==='thinking'){
+      _speakResponse();
+      return;
+    }
+    if(_origAutoRead) _origAutoRead.apply(this,arguments);
+  };
+
+  function _activate(){
+    _voiceModeActive=true;
+    modeBtn.classList.add('active');
+    modeBtn.title=t('voice_mode_active');
+    showToast(t('voice_mode_active'),1500);
+    // If the agent is busy, wait — state will be 'thinking' and we'll detect completion
+    if(typeof S!=='undefined'&&S.busy){
+      _setState('thinking');
+      return;
+    }
+    // Cancel any existing TTS
+    if(typeof stopTTS==='function') stopTTS();
+    _startListening();
+  }
+
+  function _deactivate(){
+    _voiceModeActive=false;
+    _voiceModeState='idle';
+    _voiceModeThinkingSid=null;
+    modeBtn.classList.remove('active');
+    modeBtn.title=t('voice_toggle');
+    bar.style.display='none';
+    clearTimeout(_silenceTimer);
+    try{ if(_recognition) _recognition.abort(); }catch(_){}
+    _recognition=null;
+    if(typeof stopTTS==='function') stopTTS();
+    // Restore original autoReadLastAssistant
+    if(_origAutoRead) window.autoReadLastAssistant=_origAutoRead;
+    // Clear textarea if it was only voice input
+    ta.value='';
+    autoResize();
+  }
+
+  modeBtn.onclick=()=>{
+    if(_voiceModeActive){
+      _deactivate();
+      showToast(t('voice_mode_off'),1500);
+    }else{
+      _activate();
+    }
+  };
+
+  // Expose for external use
+  window._voiceModeActive=()=>_voiceModeActive;
+  window._voiceModeDeactivate=_deactivate;
+  window._voiceModeImmediateSend=_voiceModeSend;
+})();
 $('fileInput').onchange=e=>{addFiles(Array.from(e.target.files));e.target.value='';};
 $('btnNewChat').onclick=async()=>{
   // If the current session has no messages, just focus the composer rather than
@@ -463,21 +747,31 @@ $('btnClearPreview').onclick=handleWorkspaceClose;
 $('modelSelect').onchange=async()=>{
   if(!S.session)return;
   const selectedModel=$('modelSelect').value;
+  const modelState=(typeof _modelStateForSelect==='function')
+    ? _modelStateForSelect($('modelSelect'),selectedModel)
+    : {model:selectedModel,model_provider:null};
   if(typeof closeModelDropdown==='function') closeModelDropdown();
-  localStorage.setItem('hermes-webui-model', selectedModel);
-  await api('/api/session/update',{method:'POST',body:JSON.stringify({session_id:S.session.session_id,workspace:S.session.workspace,model:selectedModel})});
-  S.session.model=selectedModel;
+  if(typeof _writePersistedModelState==='function') _writePersistedModelState(modelState.model,modelState.model_provider);
+  else localStorage.setItem('hermes-webui-model', modelState.model);
+  await api('/api/session/update',{method:'POST',body:JSON.stringify({
+    session_id:S.session.session_id,
+    workspace:S.session.workspace,
+    model:modelState.model,
+    model_provider:modelState.model_provider||null,
+  })});
+  S.session.model=modelState.model;
+  S.session.model_provider=modelState.model_provider||null;
   if(typeof syncModelChip==='function') syncModelChip();
   syncTopbar();
+  // Clarify scope: composer model changes are session-local, not the global default.
+  if(typeof showToast==='function'){
+    showToast(t('model_scope_toast')||'Applies to this conversation from your next message.', 3000);
+  }
   // Warn if selected model belongs to a different provider than what Hermes is configured for
   if(typeof _checkProviderMismatch==='function'){
     const warn=_checkProviderMismatch(selectedModel);
     if(warn&&typeof showToast==='function') showToast(warn,4000);
   }
-  // Clarify scope: composer model changes are session-local, not the global default.
-  if(typeof showToast==='function'){
-    showToast(t('model_scope_toast')||'Applies to this conversation from your next message.', 3000);
-  }
 };
 $('msg').addEventListener('input',()=>{
   autoResize();
@@ -913,11 +1207,20 @@ function applyBotName(){
   // options are enough for first paint; the dynamic provider list can settle
   // after the saved session is visible.
   const _modelDropdownReady=populateModelDropdown().then(()=>{
-    const savedModel=localStorage.getItem('hermes-webui-model');
+    const savedState=(typeof _readPersistedModelState==='function')
+      ? _readPersistedModelState()
+      : (localStorage.getItem('hermes-webui-model')?{model:localStorage.getItem('hermes-webui-model'),model_provider:null}:null);
+    const savedModel=savedState&&savedState.model;
     if(savedModel && $('modelSelect')){
-      $('modelSelect').value=savedModel;
+      const applied=(typeof _applyModelToDropdown==='function')
+        ? _applyModelToDropdown(savedModel,$('modelSelect'),savedState.model_provider||null)
+        : null;
+      if(!applied) $('modelSelect').value=savedModel;
       // If the value didn't take (model not in list), clear the bad pref
-      if($('modelSelect').value!==savedModel) localStorage.removeItem('hermes-webui-model');
+      if(!applied&&$('modelSelect').value!==savedModel){
+        if(typeof _clearPersistedModelState==='function') _clearPersistedModelState();
+        else localStorage.removeItem('hermes-webui-model');
+      }
       else if(typeof syncModelChip==='function') syncModelChip();
     }
     if(S.session) syncTopbar();
diff --git a/static/i18n.js b/static/i18n.js
index b3b21b07..73c765d2 100644
--- a/static/i18n.js
+++ b/static/i18n.js
@@ -15,6 +15,14 @@ const LOCALES = {
     mic_no_speech: 'No speech detected. Try again.',
     mic_network: 'Speech recognition unavailable.',
     mic_error: 'Voice input error: ',
+    // Turn-based voice mode (#1333)
+    voice_toggle: 'Voice input',
+    voice_listening: 'Listening…',
+    voice_speaking: 'Speaking…',
+    voice_thinking: 'Thinking…',
+    voice_error: 'Voice not supported in this browser',
+    voice_mode_active: 'Voice mode on',
+    voice_mode_off: 'Voice mode off',
     session_imported: 'Session imported',
     import_failed: 'Import failed: ',
     import_invalid_json: 'Invalid JSON',
@@ -107,6 +115,7 @@ const LOCALES = {
     model_custom_placeholder: 'e.g. openai/gpt-5.4',
     model_search_placeholder: 'Search models…',
     model_search_no_results: 'No models found',
+    model_group_configured: 'Configured',
     model_scope_advisory: 'Applies to this conversation from your next message.',
     model_scope_toast: 'Applies to this conversation from your next message.',
     // commands.js
@@ -185,6 +194,7 @@ const LOCALES = {
    branch_failed:'Fork failed: ',
    fork_from_here:'Fork from here',
    forked_from:'Forked from',
+   subagent_children:'Subagent sessions',
    btw_asking:'Asking side question...',
   btw_label:'Side question — not in history',
   btw_done:'Side question answered',
@@ -419,6 +429,7 @@ const LOCALES = {
     tab_workspaces: 'Spaces',
     tab_profiles: 'Profiles',
     tab_todos: 'Todos',
+    tab_insights: 'Insights',
     tab_settings: 'Settings',
     new_conversation: 'New conversation',
     filter_conversations: 'Filter conversations...',
@@ -439,6 +450,22 @@ const LOCALES = {
     new_skill: 'New skill',
     personal_memory: 'Personal memory',
     current_task_list: 'Current task list',
+    // Insights
+    insights_title: 'Usage Analytics',
+    insights_sessions: 'Sessions',
+    insights_messages: 'Messages',
+    insights_tokens: 'Tokens',
+    insights_cost: 'Estimated Cost',
+    insights_no_cost: 'N/A',
+    insights_models: 'Models',
+    insights_activity_by_day: 'Activity by Day',
+    insights_activity_by_hour: 'Activity by Hour',
+    insights_peak_hour: 'Peak: {hour}',
+    insights_token_breakdown: 'Token Breakdown',
+    insights_input_tokens: 'Input',
+    insights_output_tokens: 'Output',
+    insights_total: 'Total',
+    insights_footer: 'Showing data from the last {days} days',
     workspace_desc: 'Add and switch workspaces for your sessions.',
     session_meta_messages: (n) => `${n} msg${n === 1 ? '' : 's'}`,
     new_profile: 'New profile',
@@ -462,6 +489,8 @@ const LOCALES = {
     settings_label_notifications: 'Browser notifications',
     settings_desc_notifications: 'Show a system notification when a response completes while the app is in the background.',
     settings_desc_token_usage: 'Displays input/output token count below each assistant reply. Also toggled with /usage.',
+    settings_label_api_redact: 'Redact sensitive data in API responses',
+    settings_desc_api_redact: 'Self-hosted users can disable for transparency (not recommended for shared instances).',
     settings_sidebar_density_compact: 'Compact',
     settings_sidebar_density_detailed: 'Detailed',
     settings_desc_sidebar_density: 'Controls how much metadata the session list shows in the left sidebar.',
@@ -809,6 +838,22 @@ const LOCALES = {
     excalidraw_empty: 'Empty diagram',
     excalidraw_render_error: 'Failed to render diagram',
     excalidraw_simplified: 'Simplified SVG preview — not pixel-identical to Excalidraw canvas',
+    // ── Checkpoints / Rollback ──
+    checkpoint_title: 'Checkpoints',
+    checkpoint_empty: 'No checkpoints found for this workspace.',
+    checkpoint_loading: 'Loading checkpoints…',
+    checkpoint_error: 'Failed to load checkpoints',
+    checkpoint_date: 'Date',
+    checkpoint_message: 'Message',
+    checkpoint_files: 'Files',
+    checkpoint_view_diff: 'View diff',
+    checkpoint_restore: 'Restore',
+    checkpoint_restore_confirm_title: 'Restore checkpoint?',
+    checkpoint_restore_confirm_message: (ckpt) => `Restore workspace to checkpoint "${ckpt}"? This will overwrite files with the saved versions. Files added after this checkpoint will not be deleted.`,
+    checkpoint_restored: 'Checkpoint restored',
+    checkpoint_diff_title: 'Changes in checkpoint',
+    checkpoint_diff_no_changes: 'No differences found between this checkpoint and the current workspace.',
+    checkpoint_diff_files_changed: (n) => `${n} file${n === 1 ? '' : 's'} changed`,
   },
 
   ru: {
@@ -821,6 +866,13 @@ const LOCALES = {
     mic_no_speech: 'Речь не распознана. Попробуйте ещё раз.',
     mic_network: 'Распознавание речи недоступно.',
     mic_error: 'Ошибка ввода речи: ',
+    voice_toggle: 'Голосовой ввод',
+    voice_listening: 'Слушаю…',
+    voice_speaking: 'Говорю…',
+    voice_thinking: 'Думаю…',
+    voice_error: 'Голосовой ввод не поддерживается в этом браузере',
+    voice_mode_active: 'Голосовой режим включён',
+    voice_mode_off: 'Голосовой режим выключен',
     session_imported: 'Сеанс импортирован',
     import_failed: 'Не удалось импортировать: ',
     import_invalid_json: 'Неверный JSON',
@@ -960,6 +1012,7 @@ const LOCALES = {
     preserved_task_list_label: 'Сохранённый список задач',
     focus_label: 'Фокус',
     model_search_no_results: 'Модели не найдены',
+    model_group_configured: 'Настроенные',
     model_search_placeholder: 'Поиск моделей…',
     model_scope_advisory: 'Применяется к этой беседе со следующего сообщения.',
     model_scope_toast: 'Применяется к этой беседе со следующего сообщения.',
@@ -1072,6 +1125,7 @@ const LOCALES = {
     tab_workspaces: 'Рабочие пространства',
     tab_profiles: 'Профили',
     tab_todos: 'Список дел',
+    tab_insights: 'Аналитика',
     tab_settings: 'Настройки',
     new_conversation: 'Новая беседа',
     filter_conversations: 'Фильтр бесед...',
@@ -1559,6 +1613,39 @@ const LOCALES = {
     settings_label_tts_rate: 'Скорость речи',
     settings_label_tts_pitch: 'Тон речи',
 
+    checkpoint_date: 'Date',  // TODO: translate
+    checkpoint_diff_files_changed: (n) => `${n} file${n === 1 ? '' : 's'} changed`,  // TODO: translate
+    checkpoint_diff_no_changes: 'No differences found between this checkpoint and the current workspace.',  // TODO: translate
+    checkpoint_diff_title: 'Changes in checkpoint',  // TODO: translate
+    checkpoint_empty: 'No checkpoints found for this workspace.',  // TODO: translate
+    checkpoint_error: 'Failed to load checkpoints',  // TODO: translate
+    checkpoint_files: 'Files',  // TODO: translate
+    checkpoint_loading: 'Loading checkpoints…',  // TODO: translate
+    checkpoint_message: 'Message',  // TODO: translate
+    checkpoint_restore: 'Restore',  // TODO: translate
+    checkpoint_restore_confirm_message: (ckpt) => `Restore workspace to checkpoint "${ckpt}"? This will overwrite files with the saved versions. Files added after this checkpoint will not be deleted.`,  // TODO: translate
+    checkpoint_restore_confirm_title: 'Restore checkpoint?',  // TODO: translate
+    checkpoint_restored: 'Checkpoint restored',  // TODO: translate
+    checkpoint_title: 'Checkpoints',  // TODO: translate
+    checkpoint_view_diff: 'View diff',  // TODO: translate
+    insights_activity_by_day: 'Activity by Day',  // TODO: translate
+    insights_activity_by_hour: 'Activity by Hour',  // TODO: translate
+    insights_cost: 'Estimated Cost',  // TODO: translate
+    insights_footer: 'Showing data from the last {days} days',  // TODO: translate
+    insights_input_tokens: 'Input',  // TODO: translate
+    insights_messages: 'Messages',  // TODO: translate
+    insights_models: 'Models',  // TODO: translate
+    insights_no_cost: 'N/A',  // TODO: translate
+    insights_output_tokens: 'Output',  // TODO: translate
+    insights_peak_hour: 'Peak: {hour}',  // TODO: translate
+    insights_sessions: 'Sessions',  // TODO: translate
+    insights_title: 'Usage Analytics',  // TODO: translate
+    insights_token_breakdown: 'Token Breakdown',  // TODO: translate
+    insights_tokens: 'Tokens',  // TODO: translate
+    insights_total: 'Total',  // TODO: translate
+    settings_desc_api_redact: 'Self-hosted users can disable for transparency (not recommended for shared instances).',  // TODO: translate
+    settings_label_api_redact: 'Redact sensitive data in API responses',  // TODO: translate
+    subagent_children: 'Subagent sessions',  // TODO: translate
   },
 
   es: {
@@ -1652,6 +1739,7 @@ const LOCALES = {
     model_custom_placeholder: 'p. ej. openai/gpt-5.4',
     model_search_placeholder: 'Buscar modelos…',
     model_search_no_results: 'No se encontraron modelos',
+    model_group_configured: 'Configurados',
     model_scope_advisory: 'Se aplica a esta conversación desde tu próximo mensaje.',
     model_scope_toast: 'Se aplica a esta conversación desde tu próximo mensaje.',
     // commands.js
@@ -1829,6 +1917,7 @@ const LOCALES = {
     tab_workspaces: 'Espacios',
     tab_profiles: 'Perfiles',
     tab_todos: 'Todos',
+    tab_insights: 'Analíticas',
     tab_settings: 'Ajustes',
     new_conversation: 'Nueva conversación',
     filter_conversations: 'Filtrar conversaciones...',
@@ -2303,6 +2392,46 @@ const LOCALES = {
     settings_desc_tts_voice: 'Seleccionar voz para síntesis de voz',
     settings_label_tts_rate: 'Velocidad de voz',
     settings_label_tts_pitch: 'Tono de voz',
+    checkpoint_date: 'Date',  // TODO: translate
+    checkpoint_diff_files_changed: (n) => `${n} file${n === 1 ? '' : 's'} changed`,  // TODO: translate
+    checkpoint_diff_no_changes: 'No differences found between this checkpoint and the current workspace.',  // TODO: translate
+    checkpoint_diff_title: 'Changes in checkpoint',  // TODO: translate
+    checkpoint_empty: 'No checkpoints found for this workspace.',  // TODO: translate
+    checkpoint_error: 'Failed to load checkpoints',  // TODO: translate
+    checkpoint_files: 'Files',  // TODO: translate
+    checkpoint_loading: 'Loading checkpoints…',  // TODO: translate
+    checkpoint_message: 'Message',  // TODO: translate
+    checkpoint_restore: 'Restore',  // TODO: translate
+    checkpoint_restore_confirm_message: (ckpt) => `Restore workspace to checkpoint "${ckpt}"? This will overwrite files with the saved versions. Files added after this checkpoint will not be deleted.`,  // TODO: translate
+    checkpoint_restore_confirm_title: 'Restore checkpoint?',  // TODO: translate
+    checkpoint_restored: 'Checkpoint restored',  // TODO: translate
+    checkpoint_title: 'Checkpoints',  // TODO: translate
+    checkpoint_view_diff: 'View diff',  // TODO: translate
+    insights_activity_by_day: 'Activity by Day',  // TODO: translate
+    insights_activity_by_hour: 'Activity by Hour',  // TODO: translate
+    insights_cost: 'Estimated Cost',  // TODO: translate
+    insights_footer: 'Showing data from the last {days} days',  // TODO: translate
+    insights_input_tokens: 'Input',  // TODO: translate
+    insights_messages: 'Messages',  // TODO: translate
+    insights_models: 'Models',  // TODO: translate
+    insights_no_cost: 'N/A',  // TODO: translate
+    insights_output_tokens: 'Output',  // TODO: translate
+    insights_peak_hour: 'Peak: {hour}',  // TODO: translate
+    insights_sessions: 'Sessions',  // TODO: translate
+    insights_title: 'Usage Analytics',  // TODO: translate
+    insights_token_breakdown: 'Token Breakdown',  // TODO: translate
+    insights_tokens: 'Tokens',  // TODO: translate
+    insights_total: 'Total',  // TODO: translate
+    settings_desc_api_redact: 'Self-hosted users can disable for transparency (not recommended for shared instances).',  // TODO: translate
+    settings_label_api_redact: 'Redact sensitive data in API responses',  // TODO: translate
+    voice_error: 'Voice not supported in this browser',  // TODO: translate
+    voice_listening: 'Listening…',  // TODO: translate
+    voice_mode_active: 'Voice mode on',  // TODO: translate
+    voice_mode_off: 'Voice mode off',  // TODO: translate
+    voice_speaking: 'Speaking…',  // TODO: translate
+    voice_thinking: 'Thinking…',  // TODO: translate
+    voice_toggle: 'Voice input',  // TODO: translate
+    subagent_children: 'Subagent sessions',  // TODO: translate
   },
 
   de: {
@@ -2576,6 +2705,7 @@ const LOCALES = {
     tab_workspaces: 'Spaces',
     tab_profiles: 'Profile',
     tab_todos: 'Todos',
+    tab_insights: 'Statistiken',
     tab_settings: 'Einstellungen',
     new_conversation: 'Neuer Chat',
     filter_conversations: 'Chats filtern...',
@@ -2786,6 +2916,7 @@ const LOCALES = {
     model_custom_placeholder: 'z.B. openai/gpt-5.4',
     model_search_placeholder: 'Modelle suchen…',
     model_search_no_results: 'Keine Modelle gefunden',
+    model_group_configured: 'Konfiguriert',
     session_time_unknown: 'Unbekannt',
     session_time_minutes_ago: 'Vor {n} Minuten',
     session_time_hours_ago: 'Vor {n} Stunden',
@@ -3051,6 +3182,46 @@ const LOCALES = {
     settings_label_tts_rate: 'Sprechgeschwindigkeit',
     settings_label_tts_pitch: 'Tonhöhe',
 
+    checkpoint_date: 'Date',  // TODO: translate
+    checkpoint_diff_files_changed: (n) => `${n} file${n === 1 ? '' : 's'} changed`,  // TODO: translate
+    checkpoint_diff_no_changes: 'No differences found between this checkpoint and the current workspace.',  // TODO: translate
+    checkpoint_diff_title: 'Changes in checkpoint',  // TODO: translate
+    checkpoint_empty: 'No checkpoints found for this workspace.',  // TODO: translate
+    checkpoint_error: 'Failed to load checkpoints',  // TODO: translate
+    checkpoint_files: 'Files',  // TODO: translate
+    checkpoint_loading: 'Loading checkpoints…',  // TODO: translate
+    checkpoint_message: 'Message',  // TODO: translate
+    checkpoint_restore: 'Restore',  // TODO: translate
+    checkpoint_restore_confirm_message: (ckpt) => `Restore workspace to checkpoint "${ckpt}"? This will overwrite files with the saved versions. Files added after this checkpoint will not be deleted.`,  // TODO: translate
+    checkpoint_restore_confirm_title: 'Restore checkpoint?',  // TODO: translate
+    checkpoint_restored: 'Checkpoint restored',  // TODO: translate
+    checkpoint_title: 'Checkpoints',  // TODO: translate
+    checkpoint_view_diff: 'View diff',  // TODO: translate
+    insights_activity_by_day: 'Activity by Day',  // TODO: translate
+    insights_activity_by_hour: 'Activity by Hour',  // TODO: translate
+    insights_cost: 'Estimated Cost',  // TODO: translate
+    insights_footer: 'Showing data from the last {days} days',  // TODO: translate
+    insights_input_tokens: 'Input',  // TODO: translate
+    insights_messages: 'Messages',  // TODO: translate
+    insights_models: 'Models',  // TODO: translate
+    insights_no_cost: 'N/A',  // TODO: translate
+    insights_output_tokens: 'Output',  // TODO: translate
+    insights_peak_hour: 'Peak: {hour}',  // TODO: translate
+    insights_sessions: 'Sessions',  // TODO: translate
+    insights_title: 'Usage Analytics',  // TODO: translate
+    insights_token_breakdown: 'Token Breakdown',  // TODO: translate
+    insights_tokens: 'Tokens',  // TODO: translate
+    insights_total: 'Total',  // TODO: translate
+    settings_desc_api_redact: 'Self-hosted users can disable for transparency (not recommended for shared instances).',  // TODO: translate
+    settings_label_api_redact: 'Redact sensitive data in API responses',  // TODO: translate
+    voice_error: 'Voice not supported in this browser',  // TODO: translate
+    voice_listening: 'Listening…',  // TODO: translate
+    voice_mode_active: 'Voice mode on',  // TODO: translate
+    voice_mode_off: 'Voice mode off',  // TODO: translate
+    voice_speaking: 'Speaking…',  // TODO: translate
+    voice_thinking: 'Thinking…',  // TODO: translate
+    voice_toggle: 'Voice input',  // TODO: translate
+    subagent_children: 'Subagent sessions',  // TODO: translate
   },
 
   zh: {
@@ -3144,6 +3315,7 @@ const LOCALES = {
     model_custom_placeholder: '\u4f8b\u5982 openai/gpt-5.4',
     model_search_placeholder: '\u641c\u7d22\u6a21\u578b\u2026',
     model_search_no_results: '\u672a\u627e\u5230\u6a21\u578b',
+    model_group_configured: '已配置',
     model_scope_advisory: '\u4ece\u4e0b\u4e00\u6761\u6d88\u606f\u8d77\u5e94\u7528\u4e8e\u5f53\u524d\u5bf9\u8bdd\u3002',
     model_scope_toast: '\u4ece\u4e0b\u4e00\u6761\u6d88\u606f\u8d77\u5e94\u7528\u4e8e\u5f53\u524d\u5bf9\u8bdd\u3002',
     // commands.js
@@ -3318,6 +3490,7 @@ const LOCALES = {
     tab_skills: '技能',
     tab_tasks: '任务',
     tab_todos: '待办',
+    tab_insights: '统计',
     tab_workspaces: '工作区',
     tab_profiles: '配置',
     tab_settings: '设置',
@@ -3793,6 +3966,46 @@ const LOCALES = {
     settings_desc_tts_voice: '选择语音合成声音',
     settings_label_tts_rate: '语速',
     settings_label_tts_pitch: '音调',
+    checkpoint_date: 'Date',  // TODO: translate
+    checkpoint_diff_files_changed: (n) => `${n} file${n === 1 ? '' : 's'} changed`,  // TODO: translate
+    checkpoint_diff_no_changes: 'No differences found between this checkpoint and the current workspace.',  // TODO: translate
+    checkpoint_diff_title: 'Changes in checkpoint',  // TODO: translate
+    checkpoint_empty: 'No checkpoints found for this workspace.',  // TODO: translate
+    checkpoint_error: 'Failed to load checkpoints',  // TODO: translate
+    checkpoint_files: 'Files',  // TODO: translate
+    checkpoint_loading: 'Loading checkpoints…',  // TODO: translate
+    checkpoint_message: 'Message',  // TODO: translate
+    checkpoint_restore: 'Restore',  // TODO: translate
+    checkpoint_restore_confirm_message: (ckpt) => `Restore workspace to checkpoint "${ckpt}"? This will overwrite files with the saved versions. Files added after this checkpoint will not be deleted.`,  // TODO: translate
+    checkpoint_restore_confirm_title: 'Restore checkpoint?',  // TODO: translate
+    checkpoint_restored: 'Checkpoint restored',  // TODO: translate
+    checkpoint_title: 'Checkpoints',  // TODO: translate
+    checkpoint_view_diff: 'View diff',  // TODO: translate
+    insights_activity_by_day: 'Activity by Day',  // TODO: translate
+    insights_activity_by_hour: 'Activity by Hour',  // TODO: translate
+    insights_cost: 'Estimated Cost',  // TODO: translate
+    insights_footer: 'Showing data from the last {days} days',  // TODO: translate
+    insights_input_tokens: 'Input',  // TODO: translate
+    insights_messages: 'Messages',  // TODO: translate
+    insights_models: 'Models',  // TODO: translate
+    insights_no_cost: 'N/A',  // TODO: translate
+    insights_output_tokens: 'Output',  // TODO: translate
+    insights_peak_hour: 'Peak: {hour}',  // TODO: translate
+    insights_sessions: 'Sessions',  // TODO: translate
+    insights_title: 'Usage Analytics',  // TODO: translate
+    insights_token_breakdown: 'Token Breakdown',  // TODO: translate
+    insights_tokens: 'Tokens',  // TODO: translate
+    insights_total: 'Total',  // TODO: translate
+    settings_desc_api_redact: 'Self-hosted users can disable for transparency (not recommended for shared instances).',  // TODO: translate
+    settings_label_api_redact: 'Redact sensitive data in API responses',  // TODO: translate
+    voice_error: 'Voice not supported in this browser',  // TODO: translate
+    voice_listening: 'Listening…',  // TODO: translate
+    voice_mode_active: 'Voice mode on',  // TODO: translate
+    voice_mode_off: 'Voice mode off',  // TODO: translate
+    voice_speaking: 'Speaking…',  // TODO: translate
+    voice_thinking: 'Thinking…',  // TODO: translate
+    voice_toggle: 'Voice input',  // TODO: translate
+    subagent_children: 'Subagent sessions',  // TODO: translate
   },
 
   // Traditional Chinese (zh-Hant)
@@ -4093,6 +4306,7 @@ const LOCALES = {
     tab_skills: '\u6280\u80fd',
     tab_tasks: '\u4efb\u52d9',
     tab_todos: '待辦',
+    tab_insights: '統計',
     tab_workspaces: '\u5de5\u4f5c\u5340',
     new_conversation: '新對話',
     filter_conversations: '篩選對話',
@@ -4300,6 +4514,7 @@ const LOCALES = {
     model_custom_label: '\u81ea\u8a02\u6a21\u578b ID',
     model_custom_placeholder: '\u4f8b\u5982 openai/gpt-5.4',
     model_search_no_results: '\u627e\u4e0d\u5230\u6a21\u578b',
+    model_group_configured: '已設定',
     model_search_placeholder: '\u641c\u5c0b\u6a21\u578b\u2026',
     model_scope_advisory: '\u5f9e\u4e0b\u4e00\u5247\u8a0a\u606f\u8d77\u9069\u7528\u65bc\u6b64\u6703\u8a71\u3002',
     model_scope_toast: '\u5f9e\u4e0b\u4e00\u5247\u8a0a\u606f\u8d77\u9069\u7528\u65bc\u6b64\u6703\u8a71\u3002',
@@ -4643,6 +4858,46 @@ const LOCALES = {
     settings_label_tts_rate: '語速',
     settings_label_tts_pitch: '音調',
 
+    checkpoint_date: 'Date',  // TODO: translate
+    checkpoint_diff_files_changed: (n) => `${n} file${n === 1 ? '' : 's'} changed`,  // TODO: translate
+    checkpoint_diff_no_changes: 'No differences found between this checkpoint and the current workspace.',  // TODO: translate
+    checkpoint_diff_title: 'Changes in checkpoint',  // TODO: translate
+    checkpoint_empty: 'No checkpoints found for this workspace.',  // TODO: translate
+    checkpoint_error: 'Failed to load checkpoints',  // TODO: translate
+    checkpoint_files: 'Files',  // TODO: translate
+    checkpoint_loading: 'Loading checkpoints…',  // TODO: translate
+    checkpoint_message: 'Message',  // TODO: translate
+    checkpoint_restore: 'Restore',  // TODO: translate
+    checkpoint_restore_confirm_message: (ckpt) => `Restore workspace to checkpoint "${ckpt}"? This will overwrite files with the saved versions. Files added after this checkpoint will not be deleted.`,  // TODO: translate
+    checkpoint_restore_confirm_title: 'Restore checkpoint?',  // TODO: translate
+    checkpoint_restored: 'Checkpoint restored',  // TODO: translate
+    checkpoint_title: 'Checkpoints',  // TODO: translate
+    checkpoint_view_diff: 'View diff',  // TODO: translate
+    insights_activity_by_day: 'Activity by Day',  // TODO: translate
+    insights_activity_by_hour: 'Activity by Hour',  // TODO: translate
+    insights_cost: 'Estimated Cost',  // TODO: translate
+    insights_footer: 'Showing data from the last {days} days',  // TODO: translate
+    insights_input_tokens: 'Input',  // TODO: translate
+    insights_messages: 'Messages',  // TODO: translate
+    insights_models: 'Models',  // TODO: translate
+    insights_no_cost: 'N/A',  // TODO: translate
+    insights_output_tokens: 'Output',  // TODO: translate
+    insights_peak_hour: 'Peak: {hour}',  // TODO: translate
+    insights_sessions: 'Sessions',  // TODO: translate
+    insights_title: 'Usage Analytics',  // TODO: translate
+    insights_token_breakdown: 'Token Breakdown',  // TODO: translate
+    insights_tokens: 'Tokens',  // TODO: translate
+    insights_total: 'Total',  // TODO: translate
+    settings_desc_api_redact: 'Self-hosted users can disable for transparency (not recommended for shared instances).',  // TODO: translate
+    settings_label_api_redact: 'Redact sensitive data in API responses',  // TODO: translate
+    voice_error: 'Voice not supported in this browser',  // TODO: translate
+    voice_listening: 'Listening…',  // TODO: translate
+    voice_mode_active: 'Voice mode on',  // TODO: translate
+    voice_mode_off: 'Voice mode off',  // TODO: translate
+    voice_speaking: 'Speaking…',  // TODO: translate
+    voice_thinking: 'Thinking…',  // TODO: translate
+    voice_toggle: 'Voice input',  // TODO: translate
+    subagent_children: 'Subagent sessions',  // TODO: translate
   },
 
   pt: {
@@ -4711,6 +4966,7 @@ const LOCALES = {
     model_custom_placeholder: 'ex: openai/gpt-5.4',
     model_search_placeholder: 'Buscar modelos…',
     model_search_no_results: 'Nenhum modelo encontrado',
+    model_group_configured: 'Configurados',
     // commands.js
     cmd_clear: 'Limpar mensagens da conversa',
     cmd_compress: 'Comprimir manualmente o contexto (uso: /compress [tópico])',
@@ -4981,6 +5237,7 @@ const LOCALES = {
     tab_workspaces: 'Spaces',
     tab_profiles: 'Perfis',
     tab_todos: 'Todos',
+    tab_insights: 'Estatísticas',
     tab_settings: 'Configurações',
     new_conversation: 'Nova conversa',
     filter_conversations: 'Filtrar conversas...',
@@ -5305,6 +5562,46 @@ const LOCALES = {
     settings_desc_tts_voice: 'Selecionar voz para síntese de voz',
     settings_label_tts_rate: 'Velocidade da fala',
     settings_label_tts_pitch: 'Tom da fala',
+    checkpoint_date: 'Date',  // TODO: translate
+    checkpoint_diff_files_changed: (n) => `${n} file${n === 1 ? '' : 's'} changed`,  // TODO: translate
+    checkpoint_diff_no_changes: 'No differences found between this checkpoint and the current workspace.',  // TODO: translate
+    checkpoint_diff_title: 'Changes in checkpoint',  // TODO: translate
+    checkpoint_empty: 'No checkpoints found for this workspace.',  // TODO: translate
+    checkpoint_error: 'Failed to load checkpoints',  // TODO: translate
+    checkpoint_files: 'Files',  // TODO: translate
+    checkpoint_loading: 'Loading checkpoints…',  // TODO: translate
+    checkpoint_message: 'Message',  // TODO: translate
+    checkpoint_restore: 'Restore',  // TODO: translate
+    checkpoint_restore_confirm_message: (ckpt) => `Restore workspace to checkpoint "${ckpt}"? This will overwrite files with the saved versions. Files added after this checkpoint will not be deleted.`,  // TODO: translate
+    checkpoint_restore_confirm_title: 'Restore checkpoint?',  // TODO: translate
+    checkpoint_restored: 'Checkpoint restored',  // TODO: translate
+    checkpoint_title: 'Checkpoints',  // TODO: translate
+    checkpoint_view_diff: 'View diff',  // TODO: translate
+    insights_activity_by_day: 'Activity by Day',  // TODO: translate
+    insights_activity_by_hour: 'Activity by Hour',  // TODO: translate
+    insights_cost: 'Estimated Cost',  // TODO: translate
+    insights_footer: 'Showing data from the last {days} days',  // TODO: translate
+    insights_input_tokens: 'Input',  // TODO: translate
+    insights_messages: 'Messages',  // TODO: translate
+    insights_models: 'Models',  // TODO: translate
+    insights_no_cost: 'N/A',  // TODO: translate
+    insights_output_tokens: 'Output',  // TODO: translate
+    insights_peak_hour: 'Peak: {hour}',  // TODO: translate
+    insights_sessions: 'Sessions',  // TODO: translate
+    insights_title: 'Usage Analytics',  // TODO: translate
+    insights_token_breakdown: 'Token Breakdown',  // TODO: translate
+    insights_tokens: 'Tokens',  // TODO: translate
+    insights_total: 'Total',  // TODO: translate
+    settings_desc_api_redact: 'Self-hosted users can disable for transparency (not recommended for shared instances).',  // TODO: translate
+    settings_label_api_redact: 'Redact sensitive data in API responses',  // TODO: translate
+    voice_error: 'Voice not supported in this browser',  // TODO: translate
+    voice_listening: 'Listening…',  // TODO: translate
+    voice_mode_active: 'Voice mode on',  // TODO: translate
+    voice_mode_off: 'Voice mode off',  // TODO: translate
+    voice_speaking: 'Speaking…',  // TODO: translate
+    voice_thinking: 'Thinking…',  // TODO: translate
+    voice_toggle: 'Voice input',  // TODO: translate
+    subagent_children: 'Subagent sessions',  // TODO: translate
   },
   ko: {
     _lang: 'ko',
@@ -5397,6 +5694,7 @@ const LOCALES = {
     model_custom_placeholder: 'e.g. openai/gpt-5.4',
     model_search_placeholder: 'Search models…',
     model_search_no_results: 'No models found',
+    model_group_configured: '구성됨',
     model_scope_advisory: '다음 메시지부터 이 대화에 적용됩니다.',
     model_scope_toast: '다음 메시지부터 이 대화에 적용됩니다.',
     // commands.js
@@ -5694,6 +5992,7 @@ const LOCALES = {
     tab_workspaces: '공간',
     tab_profiles: 'Agent 프로필',
     tab_todos: 'Todos',
+    tab_insights: '통계',
     tab_settings: '설정',
     new_conversation: '새 대화',
     filter_conversations: '대화 필터…',
@@ -6101,6 +6400,46 @@ const LOCALES = {
     settings_desc_tts_voice: '음성 합성 음성 선택',
     settings_label_tts_rate: '말 속도',
     settings_label_tts_pitch: '말 톤',
+    checkpoint_date: 'Date',  // TODO: translate
+    checkpoint_diff_files_changed: (n) => `${n} file${n === 1 ? '' : 's'} changed`,  // TODO: translate
+    checkpoint_diff_no_changes: 'No differences found between this checkpoint and the current workspace.',  // TODO: translate
+    checkpoint_diff_title: 'Changes in checkpoint',  // TODO: translate
+    checkpoint_empty: 'No checkpoints found for this workspace.',  // TODO: translate
+    checkpoint_error: 'Failed to load checkpoints',  // TODO: translate
+    checkpoint_files: 'Files',  // TODO: translate
+    checkpoint_loading: 'Loading checkpoints…',  // TODO: translate
+    checkpoint_message: 'Message',  // TODO: translate
+    checkpoint_restore: 'Restore',  // TODO: translate
+    checkpoint_restore_confirm_message: (ckpt) => `Restore workspace to checkpoint "${ckpt}"? This will overwrite files with the saved versions. Files added after this checkpoint will not be deleted.`,  // TODO: translate
+    checkpoint_restore_confirm_title: 'Restore checkpoint?',  // TODO: translate
+    checkpoint_restored: 'Checkpoint restored',  // TODO: translate
+    checkpoint_title: 'Checkpoints',  // TODO: translate
+    checkpoint_view_diff: 'View diff',  // TODO: translate
+    insights_activity_by_day: 'Activity by Day',  // TODO: translate
+    insights_activity_by_hour: 'Activity by Hour',  // TODO: translate
+    insights_cost: 'Estimated Cost',  // TODO: translate
+    insights_footer: 'Showing data from the last {days} days',  // TODO: translate
+    insights_input_tokens: 'Input',  // TODO: translate
+    insights_messages: 'Messages',  // TODO: translate
+    insights_models: 'Models',  // TODO: translate
+    insights_no_cost: 'N/A',  // TODO: translate
+    insights_output_tokens: 'Output',  // TODO: translate
+    insights_peak_hour: 'Peak: {hour}',  // TODO: translate
+    insights_sessions: 'Sessions',  // TODO: translate
+    insights_title: 'Usage Analytics',  // TODO: translate
+    insights_token_breakdown: 'Token Breakdown',  // TODO: translate
+    insights_tokens: 'Tokens',  // TODO: translate
+    insights_total: 'Total',  // TODO: translate
+    settings_desc_api_redact: 'Self-hosted users can disable for transparency (not recommended for shared instances).',  // TODO: translate
+    settings_label_api_redact: 'Redact sensitive data in API responses',  // TODO: translate
+    voice_error: 'Voice not supported in this browser',  // TODO: translate
+    voice_listening: 'Listening…',  // TODO: translate
+    voice_mode_active: 'Voice mode on',  // TODO: translate
+    voice_mode_off: 'Voice mode off',  // TODO: translate
+    voice_speaking: 'Speaking…',  // TODO: translate
+    voice_thinking: 'Thinking…',  // TODO: translate
+    voice_toggle: 'Voice input',  // TODO: translate
+    subagent_children: 'Subagent sessions',  // TODO: translate
   },
 };
 
diff --git a/static/index.html b/static/index.html
index b89c3263..a4b15258 100644
--- a/static/index.html
+++ b/static/index.html
@@ -88,6 +88,7 @@
     <button class="rail-btn nav-tab" data-panel="workspaces" onclick="switchPanel('workspaces')" title="Spaces" data-i18n-title="tab_workspaces" aria-label="Spaces"><svg width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="1.5" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M22 19a2 2 0 0 1-2 2H4a2 2 0 0 1-2-2V5a2 2 0 0 1 2-2h5l2 3h9a2 2 0 0 1 2 2z"/></svg></button>
     <button class="rail-btn nav-tab" data-panel="profiles" onclick="switchPanel('profiles')" title="Agent profiles" data-i18n-title="tab_profiles" aria-label="Agent profiles"><svg width="18" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="1.5" stroke-linecap="round" stroke-linejoin="round"><path d="M20 21v-2a4 4 0 0 0-4-4H8a4 4 0 0 0-4 4v2"/><circle cx="12" cy="7" r="4"/></svg></button>
     <button class="rail-btn nav-tab" data-panel="todos" onclick="switchPanel('todos')" title="Current task list" data-i18n-title="tab_todos" aria-label="Todos"><svg width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="1.5" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><rect x="3" y="5" width="6" height="6" rx="1"/><path d="m3 17 2 2 4-4"/><path d="M13 6h8"/><path d="M13 12h8"/><path d="M13 18h8"/></svg></button>
+    <button class="rail-btn nav-tab" data-panel="insights" onclick="switchPanel('insights')" title="Insights" data-i18n-title="tab_insights" aria-label="Insights"><svg width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="1.5" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M18 20V10"/><path d="M12 20V4"/><path d="M6 20v-6"/></svg></button>
     <div class="rail-spacer"></div>
     <button class="rail-btn nav-tab" data-panel="settings" onclick="switchPanel('settings')" title="Settings" data-i18n-title="tab_settings" aria-label="Settings"><svg width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="1.5" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><circle cx="12" cy="12" r="3"/><path d="M19.4 15a1.65 1.65 0 0 0 .33 1.82l.06.06a2 2 0 0 1-2.83 2.83l-.06-.06a1.65 1.65 0 0 0-1.82-.33 1.65 1.65 0 0 0-1 1.51V21a2 2 0 0 1-4 0v-.09A1.65 1.65 0 0 0 9 19.4a1.65 1.65 0 0 0-1.82.33l-.06.06a2 2 0 0 1-2.83-2.83l.06-.06a1.65 1.65 0 0 0 .33-1.82 1.65 1.65 0 0 0-1.51-1H3a2 2 0 0 1 0-4h.09A1.65 1.65 0 0 0 4.6 9a1.65 1.65 0 0 0-.33-1.82l-.06-.06a2 2 0 0 1 2.83-2.83l.06.06a1.65 1.65 0 0 0 1.82.33H9a1.65 1.65 0 0 0 1-1.51V3a2 2 0 0 1 4 0v.09a1.65 1.65 0 0 0 1 1.51 1.65 1.65 0 0 0 1.82-.33l.06-.06a2 2 0 0 1 2.83 2.83l-.06.06a1.65 1.65 0 0 0-.33 1.82V9a1.65 1.65 0 0 0 1.51 1H21a2 2 0 0 1 0 4h-.09a1.65 1.65 0 0 0-1.51 1z"/></svg></button>
   </nav>
@@ -101,6 +102,7 @@
       <button class="nav-tab" data-panel="workspaces" data-label="Spaces" onclick="switchPanel('workspaces')" title="Spaces" data-i18n-title="tab_workspaces"><svg width="18" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M22 19a2 2 0 0 1-2 2H4a2 2 0 0 1-2-2V5a2 2 0 0 1 2-2h5l2 3h9a2 2 0 0 1 2 2z"/></svg></button>
       <button class="nav-tab" data-panel="profiles" data-label="Profiles" onclick="switchPanel('profiles')" title="Agent profiles" data-i18n-title="tab_profiles"><svg width="16" height="16" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M20 21v-2a4 4 0 0 0-4-4H8a4 4 0 0 0-4 4v2"/><circle cx="12" cy="7" r="4"/></svg></button>
       <button class="nav-tab" data-panel="todos" data-label="Todos" onclick="switchPanel('todos')" title="Current task list" data-i18n-title="tab_todos"><svg width="18" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><rect x="3" y="5" width="6" height="6" rx="1"/><path d="m3 17 2 2 4-4"/><path d="M13 6h8"/><path d="M13 12h8"/><path d="M13 18h8"/></svg></button>
+      <button class="nav-tab" data-panel="insights" data-label="Insights" onclick="switchPanel('insights')" title="Insights" data-i18n-title="tab_insights"><svg width="18" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M18 20V10"/><path d="M12 20V4"/><path d="M6 20v-6"/></svg></button>
       <!-- Settings button mirrored here for mobile (rail is desktop-only via @media >=768px). Keep in sync with rail entry. -->
       <button class="nav-tab" data-panel="settings" onclick="switchPanel('settings')" title="Settings" data-i18n-title="tab_settings"><svg width="18" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><circle cx="12" cy="12" r="3"/><path d="M19.4 15a1.65 1.65 0 0 0 .33 1.82l.06.06a2 2 0 0 1-2.83 2.83l-.06-.06a1.65 1.65 0 0 0-1.82-.33 1.65 1.65 0 0 0-1 1.51V21a2 2 0 0 1-4 0v-.09A1.65 1.65 0 0 0 9 19.4a1.65 1.65 0 0 0-1.82.33l-.06.06a2 2 0 0 1-2.83-2.83l.06-.06a1.65 1.65 0 0 0 .33-1.82 1.65 1.65 0 0 0-1.51-1H3a2 2 0 0 1 0-4h.09A1.65 1.65 0 0 0 4.6 9a1.65 1.65 0 0 0-.33-1.82l-.06-.06a2 2 0 0 1 2.83-2.83l.06.06a1.65 1.65 0 0 0 1.82.33H9a1.65 1.65 0 0 0 1-1.51V3a2 2 0 0 1 4 0v.09a1.65 1.65 0 0 0 1 1.51 1.65 1.65 0 0 0 1.82-.33l.06-.06a2 2 0 0 1 2.83 2.83l-.06.06a1.65 1.65 0 0 0-.33 1.82V9a1.65 1.65 0 0 0 1.51 1H21a2 2 0 0 1 0 4h-.09a1.65 1.65 0 0 0-1.51 1z"/></svg></button>
     </div>
@@ -153,6 +155,23 @@
       </div>
       <div id="todoPanel" style="flex:1;overflow-y:auto;padding:8px 12px"></div>
     </div>
+    <!-- Insights panel -->
+    <div class="panel-view" id="panelInsights">
+      <div class="panel-head">
+        <span data-i18n="tab_insights">Insights</span>
+        <div class="panel-head-actions">
+          <button class="panel-head-btn" id="insightsRefreshBtn" onclick="loadInsights(true)" title="Refresh" aria-label="Refresh"><svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><polyline points="23 4 23 10 17 10"/><path d="M20.49 15a9 9 0 1 1-2.12-9.36L23 10"/></svg></button>
+        </div>
+      </div>
+      <div class="panel-head-sub" style="padding:0 12px 8px">
+        <select id="insightsPeriod" onchange="loadInsights()" style="width:100%;background:var(--input-bg);color:var(--text);border:1px solid var(--border);border-radius:6px;padding:4px 8px;font-size:12px">
+          <option value="7">7 days</option>
+          <option value="30" selected>30 days</option>
+          <option value="90">90 days</option>
+          <option value="365">365 days</option>
+        </select>
+      </div>
+    </div>
     <!-- Workspaces panel -->
     <div class="panel-view" id="panelWorkspaces">
       <div class="panel-head">
@@ -359,6 +378,10 @@
         </div>
         <div class="attach-tray" id="attachTray"></div>
         <div class="mic-status" id="micStatus" style="display:none"><span class="mic-dot"></span> Listening…</div>
+        <div class="voice-mode-bar" id="voiceModeBar" style="display:none">
+          <span class="voice-mode-indicator" id="voiceModeIndicator"></span>
+          <span class="voice-mode-label" id="voiceModeLabel"></span>
+        </div>
         <textarea id="msg" rows="1" placeholder="Message Hermes…"></textarea>
         <div class="composer-footer">
           <div class="composer-left">
@@ -374,6 +397,16 @@
                 <line x1="8" y1="23" x2="16" y2="23"/>
               </svg>
             </button>
+            <button class="icon-btn voice-mode-btn" id="btnVoiceMode" title="Turn-based voice mode" style="display:none" data-i18n-title="voice_toggle">
+              <svg width="16" height="16" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+                <path d="M12 1a3 3 0 0 0-3 3v8a3 3 0 0 0 6 0V4a3 3 0 0 0-3-3z"/>
+                <path d="M19 10v2a7 7 0 0 1-14 0v-2"/>
+                <line x1="12" y1="19" x2="12" y2="23"/>
+                <line x1="8" y1="23" x2="16" y2="23"/>
+                <path d="M20 3l-1.5 1.5" opacity=".5"/>
+                <path d="M4 3l1.5 1.5" opacity=".5"/>
+              </svg>
+            </button>
             <div class="composer-divider" aria-hidden="true"></div>
             <button class="yolo-pill" id="yoloPill" type="button" onclick="cmdYolo()" style="display:none" title="YOLO mode — click to disable" data-i18n-title="yolo_pill_title_active">
               <span class="yolo-pill-icon" aria-hidden="true">⚡</span>
@@ -594,6 +627,14 @@
         <div class="main-view-empty-sub" data-i18n="profiles_empty_sub">Pick an agent profile from the sidebar to view and edit its settings, or create a new one.</div>
       </div>
     </div>
+    <div id="mainInsights" class="main-view">
+      <div class="main-view-header">
+        <div class="main-view-title" data-i18n="insights_title">Usage Analytics</div>
+      </div>
+      <div class="main-view-content" id="insightsContent" style="padding:16px;overflow-y:auto">
+        <div style="color:var(--muted);font-size:12px" data-i18n="loading">Loading...</div>
+      </div>
+    </div>
     <div id="mainSettings" class="main-view">
       <div class="settings-main">
           <div class="settings-pane active" id="settingsPaneConversation">
@@ -767,6 +808,13 @@
               </label>
               <div style="font-size:11px;color:var(--muted);margin-top:4px">Group thinking and tool calls into one collapsed activity section per assistant turn.</div>
             </div>
+            <div class="settings-field">
+              <label style="display:flex;align-items:center;gap:8px;cursor:pointer">
+                <input type="checkbox" id="settingsApiRedact" checked style="width:15px;height:15px;accent-color:var(--accent)">
+                <span data-i18n="settings_label_api_redact">Redact sensitive data in API responses</span>
+              </label>
+              <div style="font-size:11px;color:var(--muted);margin-top:4px" data-i18n="settings_desc_api_redact">Self-hosted users can disable for transparency (not recommended for shared instances).</div>
+            </div>
             <div class="settings-field">
               <label for="settingsSidebarDensity" data-i18n="settings_label_sidebar_density">Sidebar density</label>
               <select id="settingsSidebarDensity" style="width:100%;padding:8px;background:var(--code-bg);color:var(--text);border:1px solid var(--border2);border-radius:6px">
diff --git a/static/messages.js b/static/messages.js
index d9ea8fba..ceb3b475 100644
--- a/static/messages.js
+++ b/static/messages.js
@@ -82,7 +82,7 @@ async function send(){
         S.pendingFiles=[];renderTray();
       } else if(busyMode==='interrupt'){
         // Queue the message, then cancel so drain re-sends it.
-        queueSessionMessage(S.session.session_id,{text,files:[...S.pendingFiles],model:S.session&&S.session.model||($('modelSelect')&&$('modelSelect').value)||'',profile:S.activeProfile||'default'});
+        queueSessionMessage(S.session.session_id,{text,files:[...S.pendingFiles],model:S.session&&S.session.model||($('modelSelect')&&$('modelSelect').value)||'',model_provider:S.session&&S.session.model_provider||null,profile:S.activeProfile||'default'});
         updateQueueBadge(S.session.session_id);
         $('msg').value='';autoResize();
         S.pendingFiles=[];renderTray();
@@ -95,7 +95,7 @@ async function send(){
       } else {
         // Default: queue mode (current behavior). Also the fallback for
         // 'steer' mode when no stream is active or _trySteer is unavailable.
-        queueSessionMessage(S.session.session_id,{text,files:[...S.pendingFiles],model:S.session&&S.session.model||($('modelSelect')&&$('modelSelect').value)||'',profile:S.activeProfile||'default'});
+        queueSessionMessage(S.session.session_id,{text,files:[...S.pendingFiles],model:S.session&&S.session.model||($('modelSelect')&&$('modelSelect').value)||'',model_provider:S.session&&S.session.model_provider||null,profile:S.activeProfile||'default'});
         $('msg').value='';autoResize();
         S.pendingFiles=[];renderTray();
         updateQueueBadge(S.session.session_id);
@@ -202,12 +202,21 @@ async function send(){
     const startData=await api('/api/chat/start',{method:'POST',body:JSON.stringify({
       session_id:activeSid,message:msgText,
       model:S.session.model||$('modelSelect').value,workspace:S.session.workspace,
+      model_provider:S.session.model_provider||null,
       attachments:uploaded.length?uploaded:undefined
     })});
     if(startData.effective_model && S.session){
       S.session.model=startData.effective_model;
+      S.session.model_provider=startData.effective_model_provider||S.session.model_provider||null;
       localStorage.setItem('hermes-webui-model', startData.effective_model);
-      if($('modelSelect')) _applyModelToDropdown(startData.effective_model, $('modelSelect'));
+      if(typeof _writePersistedModelState==='function') _writePersistedModelState(startData.effective_model,S.session.model_provider||null);
+      if($('modelSelect')) _applyModelToDropdown(startData.effective_model, $('modelSelect'),S.session.model_provider||null);
+      if(typeof syncTopbar==='function') syncTopbar();
+    }else if(startData.effective_model_provider && S.session){
+      S.session.model_provider=startData.effective_model_provider;
+      if(typeof _writePersistedModelState==='function') _writePersistedModelState(S.session.model||'',S.session.model_provider||null);
+      if($('modelSelect')&&typeof _applyModelToDropdown==='function') _applyModelToDropdown(S.session.model||'', $('modelSelect'), S.session.model_provider||null);
+      if(typeof syncModelChip==='function') syncModelChip();
       if(typeof syncTopbar==='function') syncTopbar();
     }
     streamId=startData.stream_id;
@@ -233,7 +242,7 @@ async function send(){
       stopApprovalPolling();
       stopClarifyPolling();
       // Keep the user's attempted turn by queueing it for after the current run.
-      queueSessionMessage(activeSid,{text:msgText,files:[],model:S.session&&S.session.model||($('modelSelect')&&$('modelSelect').value)||'',profile:S.activeProfile||'default'});
+      queueSessionMessage(activeSid,{text:msgText,files:[],model:S.session&&S.session.model||($('modelSelect')&&$('modelSelect').value)||'',model_provider:S.session&&S.session.model_provider||null,profile:S.activeProfile||'default'});
       updateQueueBadge(activeSid);
       showToast('Current session is still running. Reconnected and queued your message.',2600);
       try{
@@ -898,6 +907,7 @@ function attachLiveStream(activeSid, streamId, uploaded=[], options={}){
           queueSessionMessage(sid,{
             text:txt,files:[],
             model:S.session&&S.session.model||'',
+            model_provider:S.session&&S.session.model_provider||null,
             profile:S.activeProfile||'default',
           });
           if(typeof updateQueueBadge==='function') updateQueueBadge(sid);
diff --git a/static/panels.js b/static/panels.js
index b49fe262..2d325edc 100644
--- a/static/panels.js
+++ b/static/panels.js
@@ -160,11 +160,11 @@ async function switchPanel(name, opts = {}) {
   document.querySelectorAll('.panel-view').forEach(p => p.classList.remove('active'));
   const panelEl = $('panel' + nextPanel.charAt(0).toUpperCase() + nextPanel.slice(1));
   if (panelEl) panelEl.classList.add('active');
-  // Toggle main content view. Each entry in MAIN_VIEW_PANELS gets a matching
+  // Update main content view. Each entry in MAIN_VIEW_PANELS gets a matching
   // showing-<name> class on <main>; no class means chat (the default).
   const mainEl = document.querySelector('main.main');
   if (mainEl) {
-    ['settings','skills','memory','tasks','workspaces','profiles'].forEach(p => {
+    ['settings','skills','memory','tasks','workspaces','profiles','insights'].forEach(p => {
       mainEl.classList.toggle('showing-' + p, nextPanel === p);
     });
   }
@@ -175,6 +175,7 @@ async function switchPanel(name, opts = {}) {
   if (nextPanel === 'workspaces') await loadWorkspacesPanel();
   if (nextPanel === 'profiles') await loadProfilesPanel();
   if (nextPanel === 'todos') loadTodos();
+  if (nextPanel === 'insights') await loadInsights();
   if (nextPanel === 'settings') {
     switchSettingsSection(_currentSettingsSection);
     loadSettingsPanel();
@@ -848,6 +849,112 @@ function loadTodos() {
     </div>`).join('');
 }
 
+// ── Insights panel ──
+async function loadInsights(animate) {
+  const box = $('insightsContent');
+  const refreshBtn = $('insightsRefreshBtn');
+  if (!box) return;
+  if (animate && refreshBtn) {
+    refreshBtn.style.opacity = '0.5';
+    refreshBtn.disabled = true;
+  }
+  const period = ($('insightsPeriod') || {}).value || '30';
+  try {
+    const data = await api(`/api/insights?days=${period}`);
+    _renderInsights(data, box);
+  } catch(e) {
+    box.innerHTML = `<div style="color:var(--accent);font-size:12px">${esc(t('error_prefix') + e.message)}</div>`;
+  } finally {
+    if (animate && refreshBtn) {
+      refreshBtn.style.opacity = '';
+      refreshBtn.disabled = false;
+    }
+  }
+}
+
+function _renderInsights(d, box) {
+  const fmtNum = n => n.toLocaleString();
+  const fmtCost = c => c > 0 ? '$' + c.toFixed(4) : t('insights_no_cost');
+  const fmtTokens = n => n >= 1e6 ? (n/1e6).toFixed(1) + 'M' : n >= 1e3 ? (n/1e3).toFixed(1) + 'K' : fmtNum(n);
+
+  // Overview cards
+  const overviewCards = [
+    { label: t('insights_sessions'), value: fmtNum(d.total_sessions), icon: li('message-square', 18) },
+    { label: t('insights_messages'), value: fmtNum(d.total_messages), icon: li('hash', 18) },
+    { label: t('insights_tokens'), value: fmtTokens(d.total_tokens), icon: li('cpu', 18) },
+    { label: t('insights_cost'), value: fmtCost(d.total_cost), icon: li('dollar-sign', 18) },
+  ];
+
+  // Models table
+  let modelsHtml = '';
+  if (d.models && d.models.length) {
+    const totalSess = d.models.reduce((a, m) => a + m.sessions, 0) || 1;
+    modelsHtml = `<div class="insights-card"><div class="insights-card-title">${esc(t('insights_models'))}</div><div class="insights-table"><div class="insights-table-head"><span>Model</span><span>Sessions</span><span>Share</span></div>` +
+      d.models.map(m => {
+        const pct = ((m.sessions / totalSess) * 100).toFixed(0);
+        return `<div class="insights-table-row"><span class="insights-model-name" title="${esc(m.model)}">${esc(m.model)}</span><span>${m.sessions}</span><span>${pct}%</span></div>`;
+      }).join('') +
+      `</div></div>`;
+  }
+
+  // Activity by day of week
+  let dowHtml = '';
+  if (d.activity_by_day) {
+    const maxDow = Math.max(...d.activity_by_day.map(x => x.sessions), 1);
+    dowHtml = `<div class="insights-card"><div class="insights-card-title">${esc(t('insights_activity_by_day'))}</div><div class="insights-bars">` +
+      d.activity_by_day.map(r => {
+        const pct = (r.sessions / maxDow * 100).toFixed(0);
+        return `<div class="insights-bar-row"><span class="insights-bar-label">${r.day}</span><div class="insights-bar-track"><div class="insights-bar-fill" style="width:${pct}%"></div></div><span class="insights-bar-value">${r.sessions}</span></div>`;
+      }).join('') +
+      `</div></div>`;
+  }
+
+  // Activity by hour
+  let hodHtml = '';
+  if (d.activity_by_hour) {
+    const maxHod = Math.max(...d.activity_by_hour.map(x => x.sessions), 1);
+    const peakHour = d.activity_by_hour.reduce((a, b) => b.sessions > a.sessions ? b : a, {hour:0,sessions:0});
+    hodHtml = `<div class="insights-card"><div class="insights-card-title">${esc(t('insights_activity_by_hour'))} <span style="font-weight:400;font-size:11px;color:var(--muted)">${esc(t('insights_peak_hour').replace('{hour}', peakHour.hour + ':00'))}</span></div><div class="insights-bars">` +
+      d.activity_by_hour.map(r => {
+        const pct = (r.sessions / maxHod * 100).toFixed(0);
+        const isPeak = r.hour === peakHour.hour && peakHour.sessions > 0;
+        return `<div class="insights-bar-row"><span class="insights-bar-label">${String(r.hour).padStart(2,'0')}</span><div class="insights-bar-track"><div class="insights-bar-fill${isPeak ? ' insights-bar-peak' : ''}" style="width:${pct}%"></div></div><span class="insights-bar-value">${r.sessions}</span></div>`;
+      }).join('') +
+      `</div></div>`;
+  }
+
+  // Token breakdown
+  const tokenCards = `
+    <div class="insights-card">
+      <div class="insights-card-title">${esc(t('insights_token_breakdown'))}</div>
+      <div class="insights-token-row">
+        <span class="insights-token-label">${esc(t('insights_input_tokens'))}</span>
+        <span class="insights-token-value">${fmtTokens(d.total_input_tokens)}</span>
+      </div>
+      <div class="insights-token-row">
+        <span class="insights-token-label">${esc(t('insights_output_tokens'))}</span>
+        <span class="insights-token-value">${fmtTokens(d.total_output_tokens)}</span>
+      </div>
+      <div class="insights-token-row insights-token-total">
+        <span class="insights-token-label">${esc(t('insights_total'))}</span>
+        <span class="insights-token-value">${fmtTokens(d.total_tokens)}</span>
+      </div>
+    </div>`;
+
+  box.innerHTML = `
+    <div class="insights-grid">
+      ${overviewCards.map(c => `<div class="insights-stat"><div class="insights-stat-icon">${c.icon}</div><div class="insights-stat-info"><div class="insights-stat-value">${c.value}</div><div class="insights-stat-label">${esc(c.label)}</div></div></div>`).join('')}
+    </div>
+    <div class="insights-row">
+      ${tokenCards}
+      ${modelsHtml}
+    </div>
+    ${dowHtml}
+    ${hodHtml}
+    <div style="text-align:center;color:var(--muted);font-size:10px;margin-top:12px;opacity:.6">${esc(t('insights_footer').replace('{days}', d.period_days))}</div>
+  `;
+}
+
 async function clearConversation() {
   if(!S.session) return;
   const _clrMsg=await showConfirmDialog({title:t('clear_conversation_title'),message:t('clear_conversation_message'),confirmLabel:t('clear'),danger:true,focusCancel:true});
@@ -1698,11 +1805,18 @@ function _renderWorkspaceDetail(ws){
         <div class="detail-row"><div class="detail-row-label">Path</div><div class="detail-row-value"><code>${esc(ws.path)}</code></div></div>
         <div class="detail-row"><div class="detail-row-label">Status</div><div class="detail-row-value">${statusBadge}${defaultBadge}</div></div>
       </div>
+      <div class="detail-card" style="margin-top:12px">
+        <div class="detail-card-title">${esc(t('checkpoint_title'))}</div>
+        <div id="checkpointListContainer">
+          <div style="color:var(--muted);font-size:12px;padding:8px 0">${esc(t('checkpoint_loading'))}</div>
+        </div>
+      </div>
     </div>`;
   body.style.display = '';
   if (empty) empty.style.display = 'none';
   _workspaceMode = 'read';
   _setWorkspaceHeaderButtons('read', ws);
+  _loadCheckpoints(ws.path);
 }
 
 function _setWorkspaceHeaderButtons(mode, ws){
@@ -2002,7 +2116,7 @@ async function switchToWorkspace(path,name){
   try{
     closeWsDropdown();
     await api('/api/session/update',{method:'POST',body:JSON.stringify({
-      session_id:S.session.session_id, workspace:path, model:S.session.model
+      session_id:S.session.session_id, workspace:path, model:S.session.model, model_provider:S.session.model_provider||null
     })});
     S.session.workspace=path;
     // Explicit workspace switch = user overriding any pending profile-switch default.
@@ -2252,10 +2366,15 @@ async function switchToProfile(name) {
     const data = await api('/api/profile/switch', { method: 'POST', body: JSON.stringify({ name }) });
     S.activeProfile = data.active || name;
 
+    // Update composer placeholder and title bar while the core profile-switch
+    // state is still close to the profile API response.
+    if (typeof applyBotName === 'function') applyBotName();
+
     // ── Model + Workspace (parallelized) ───────────────────────────────────
     // populateModelDropdown hits /api/models; loadWorkspaceList hits /api/workspaces.
     // They are fully independent — run both simultaneously to cut switch time ~50%.
-    localStorage.removeItem('hermes-webui-model');
+    if(typeof _clearPersistedModelState==='function') _clearPersistedModelState();
+    else localStorage.removeItem('hermes-webui-model');
     _skillsData = null;
     _workspaceList = null;
     await Promise.all([populateModelDropdown(), loadWorkspaceList()]);
@@ -2265,10 +2384,15 @@ async function switchToProfile(name) {
       const sel = $('modelSelect');
       const resolved = _applyModelToDropdown(data.default_model, sel, window._activeProvider||null);
       const modelToUse = resolved || data.default_model;
+      const modelState = (typeof _modelStateForSelect==='function')
+        ? _modelStateForSelect(sel, modelToUse)
+        : {model:modelToUse,model_provider:null};
       S._pendingProfileModel = modelToUse;
+      S._pendingProfileModelProvider = modelState.model_provider||null;
       // Only patch the in-memory session model if we're NOT about to replace the session
       if (S.session && !sessionInProgress) {
         S.session.model = modelToUse;
+        S.session.model_provider = modelState.model_provider||null;
       }
     }
 
@@ -2288,6 +2412,7 @@ async function switchToProfile(name) {
             session_id: S.session.session_id,
             workspace: data.default_workspace,
             model: S.session.model,
+            model_provider: S.session.model_provider||null,
           })});
           S.session.workspace = data.default_workspace;
         } catch (_) {}
@@ -2308,6 +2433,7 @@ async function switchToProfile(name) {
             session_id: S.session.session_id,
             workspace: S._profileDefaultWorkspace,
             model: S.session.model,
+            model_provider: S.session.model_provider||null,
           })});
           S.session.workspace = S._profileDefaultWorkspace;
         } catch (_) {}
@@ -2334,9 +2460,6 @@ async function switchToProfile(name) {
     if (_currentPanel === 'profiles') await loadProfilesPanel();
     if (_currentPanel === 'workspaces') await loadWorkspacesPanel();
 
-    // Update composer placeholder and title bar to reflect profile name
-    if (typeof applyBotName === 'function') applyBotName();
-
   } catch (e) {
     // Revert the optimistic name update on error
     if (_chipLabel) _chipLabel.textContent = _prevProfileName;
@@ -2624,11 +2747,13 @@ function _markSettingsDirty(){
   _settingsDirty = true;
 }
 
-// Apply TTS enabled state: show/hide TTS buttons on all assistant messages
+// Apply TTS enabled state: toggles a body class so the CSS rule
+// `body.tts-enabled .msg-tts-btn` shows/hides the speaker icon. We toggle the
+// body class instead of writing inline `style.display` because the parent
+// `.msg-action-btn` has no display rule, so clearing the inline style let the
+// `.msg-tts-btn{display:none;}` cascade re-hide the button (#1409).
 function _applyTtsEnabled(enabled){
-  document.querySelectorAll('.msg-tts-btn').forEach(btn=>{
-    btn.style.display=enabled?'':'none';
-  });
+  document.body.classList.toggle('tts-enabled', !!enabled);
 }
 
 function _appearancePayloadFromUi(){
@@ -2708,6 +2833,8 @@ function _preferencesPayloadFromUi(){
   if(showUsageCb) payload.show_token_usage=showUsageCb.checked;
   const simplifiedToolCb=$('settingsSimplifiedToolCalling');
   if(simplifiedToolCb) payload.simplified_tool_calling=simplifiedToolCb.checked;
+  const apiRedactCb=$('settingsApiRedact');
+  if(apiRedactCb) payload.api_redact_enabled=apiRedactCb.checked;
   const showCliCb=$('settingsShowCliSessions');
   if(showCliCb) payload.show_cli_sessions=showCliCb.checked;
   const syncCb=$('settingsSyncInsights');
@@ -2900,6 +3027,8 @@ async function loadSettingsPanel(){
     if(showUsageCb){showUsageCb.checked=!!settings.show_token_usage;showUsageCb.addEventListener('change',_schedulePreferencesAutosave,{once:false});}
     const simplifiedToolCb=$('settingsSimplifiedToolCalling');
     if(simplifiedToolCb){simplifiedToolCb.checked=settings.simplified_tool_calling!==false;simplifiedToolCb.addEventListener('change',_schedulePreferencesAutosave,{once:false});}
+    const apiRedactCb=$('settingsApiRedact');
+    if(apiRedactCb){apiRedactCb.checked=settings.api_redact_enabled!==false;apiRedactCb.addEventListener('change',_schedulePreferencesAutosave,{once:false});}
     const showCliCb=$('settingsShowCliSessions');
     if(showCliCb){showCliCb.checked=!!settings.show_cli_sessions;showCliCb.addEventListener('change',_schedulePreferencesAutosave,{once:false});}
     const syncCb=$('settingsSyncInsights');
@@ -3344,6 +3473,7 @@ async function saveSettings(andClose){
   body.language=language;
   body.show_token_usage=showTokenUsage;
   body.simplified_tool_calling=!!($('settingsSimplifiedToolCalling')||{}).checked;
+  body.api_redact_enabled=!!($('settingsApiRedact')||{}).checked;
   body.show_cli_sessions=showCliSessions;
   body.sync_to_insights=!!($('settingsSyncInsights')||{}).checked;
   body.check_for_updates=!!($('settingsCheckUpdates')||{}).checked;
@@ -3631,3 +3761,111 @@ switchSettingsSection=function(name){
   _origSwitchSettings(name);
   if(name==='system') loadMcpServers();
 };
+
+// ── Checkpoints / Rollback ──────────────────────────────────────────────────
+
+async function _loadCheckpoints(workspace){
+  const container=$('checkpointListContainer');
+  if(!container) return;
+  try{
+    const data=await api(`/api/rollback/list?workspace=${encodeURIComponent(workspace)}`);
+    const checkpoints=data.checkpoints||[];
+    if(!checkpoints.length){
+      container.innerHTML=`<div style="color:var(--muted);font-size:12px;padding:8px 0">${esc(t('checkpoint_empty'))}</div>`;
+      return;
+    }
+    let html='';
+    for(const ck of checkpoints){
+      const shortId=ck.id||ck.commit||'?';
+      const msg=ck.message||'checkpoint';
+      const date=ck.date_display||ck.date||'';
+      const files=ck.files||0;
+      html+=`
+        <div class="detail-row" style="align-items:center;padding:6px 0;border-bottom:1px solid var(--border,rgba(255,255,255,0.08))">
+          <div style="flex:1;min-width:0">
+            <div style="font-size:13px;font-weight:500;overflow:hidden;text-overflow:ellipsis;white-space:nowrap" title="${esc(msg)}">${esc(msg)}</div>
+            <div style="font-size:11px;color:var(--muted);margin-top:2px">
+              <code style="font-size:10px">${esc(shortId)}</code>
+              ${date ? ` · ${esc(date)}` : ''}
+              ${files ? ` · ${esc(t('checkpoint_files'))}: ${files}` : ''}
+            </div>
+          </div>
+          <div style="display:flex;gap:4px;flex-shrink:0;margin-left:8px">
+            <button class="panel-head-btn" title="${esc(t('checkpoint_view_diff'))}" onclick="event.stopPropagation();_viewCheckpointDiff('${esc(workspace)}','${esc(ck.id)}')">
+              <svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" width="16" height="16"><path d="M12 20h9"/><path d="M16.5 3.5a2.121 2.121 0 0 1 3 3L7 19l-4 1 1-4L16.5 3.5z"/></svg>
+            </button>
+            <button class="panel-head-btn" title="${esc(t('checkpoint_restore'))}" onclick="event.stopPropagation();_restoreCheckpoint('${esc(workspace)}','${esc(ck.id)}','${esc(msg.replace(/'/g,"\\'"))}')">
+              <svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" width="16" height="16"><polyline points="1 4 1 10 7 10"/><path d="M3.51 15a9 9 0 1 0 2.13-9.36L1 10"/></svg>
+            </button>
+          </div>
+        </div>`;
+    }
+    container.innerHTML=html;
+  }catch(e){
+    container.innerHTML=`<div style="color:var(--error,#f87171);font-size:12px;padding:8px 0">${esc(t('checkpoint_error'))}: ${esc(e.message)}</div>`;
+  }
+}
+
+async function _viewCheckpointDiff(workspace,checkpoint){
+  const modal=document.getElementById('checkpointDiffModal');
+  if(!modal){
+    const m=document.createElement('div');
+    m.id='checkpointDiffModal';
+    m.style.cssText='position:fixed;inset:0;z-index:9999;display:flex;align-items:center;justify-content:center;background:rgba(0,0,0,0.6)';
+    m.innerHTML=`
+      <div style="background:var(--bg,${getComputedStyle(document.documentElement).getPropertyValue('--bg')||'#1a1a2e'});border:1px solid var(--border,rgba(255,255,255,0.12));border-radius:12px;width:90vw;max-width:800px;max-height:80vh;display:flex;flex-direction:column;box-shadow:0 8px 32px rgba(0,0,0,0.4)">
+        <div style="display:flex;align-items:center;justify-content:space-between;padding:12px 16px;border-bottom:1px solid var(--border,rgba(255,255,255,0.08))">
+          <div id="checkpointDiffModalTitle" style="font-weight:600;font-size:14px"></div>
+          <button onclick="document.getElementById('checkpointDiffModal').style.display='none'" style="background:none;border:none;color:var(--fg);cursor:pointer;font-size:18px;padding:0 4px">&times;</button>
+        </div>
+        <div id="checkpointDiffModalBody" style="flex:1;overflow:auto;padding:12px 16px">
+          <div style="color:var(--muted);font-size:12px">${esc(t('checkpoint_loading'))}</div>
+        </div>
+      </div>`;
+    m.onclick=(e)=>{if(e.target===m) m.style.display='none';};
+    document.body.appendChild(m);
+  }
+  modal.style.display='flex';
+  $('checkpointDiffModalTitle').textContent=t('checkpoint_diff_title');
+  $('checkpointDiffModalBody').innerHTML=`<div style="color:var(--muted);font-size:12px">${esc(t('checkpoint_loading'))}</div>`;
+  try{
+    const data=await api(`/api/rollback/diff?workspace=${encodeURIComponent(workspace)}&checkpoint=${encodeURIComponent(checkpoint)}`);
+    const body=$('checkpointDiffModalBody');
+    if(!data.total_changes){
+      body.innerHTML=`<div style="color:var(--muted);font-size:12px">${esc(t('checkpoint_diff_no_changes'))}</div>`;
+      return;
+    }
+    let html=`<div style="font-size:12px;margin-bottom:8px">${esc(t('checkpoint_diff_files_changed',data.total_changes))}</div>`;
+    if(data.files_changed){
+      html+='<div style="margin-bottom:8px">';
+      for(const f of data.files_changed){
+        const icon=f.status==='deleted'?'−':'~';
+        const color=f.status==='deleted'?'var(--error,#f87171)':'var(--accent,#60a5fa)';
+        html+=`<div style="font-size:12px;padding:2px 0"><span style="color:${color};font-weight:bold;margin-right:6px">${icon}</span><code style="font-size:11px">${esc(f.file)}</code></div>`;
+      }
+      html+='</div>';
+    }
+    if(data.diff){
+      html+=`<pre style="background:var(--bg-secondary,rgba(0,0,0,0.3));border:1px solid var(--border,rgba(255,255,255,0.08));border-radius:8px;padding:12px;font-size:11px;line-height:1.4;overflow-x:auto;white-space:pre-wrap;word-break:break-all;max-height:50vh;overflow-y:auto;color:var(--fg)">${esc(data.diff)}</pre>`;
+    }
+    body.innerHTML=html;
+  }catch(e){
+    $('checkpointDiffModalBody').innerHTML=`<div style="color:var(--error,#f87171);font-size:12px">${esc(e.message)}</div>`;
+  }
+}
+
+async function _restoreCheckpoint(workspace,checkpoint,message){
+  const label=message||checkpoint;
+  const ok=await showConfirmDialog({title:t('checkpoint_restore_confirm_title'),message:t('checkpoint_restore_confirm_message',label),confirmLabel:t('checkpoint_restore'),danger:true,focusCancel:true});
+  if(!ok) return;
+  try{
+    const data=await api('/api/rollback/restore',{method:'POST',body:JSON.stringify({workspace,checkpoint})});
+    if(data&&data.ok){
+      showToast(t('checkpoint_restored')+(data.files_restored_count?` (${data.files_restored_count} ${t('checkpoint_files').toLowerCase()})`:''));
+    }else{
+      showToast((data&&data.error)||'Restore failed','error');
+    }
+  }catch(e){
+    showToast(t('checkpoint_restore')+': '+e.message,'error');
+  }
+}
diff --git a/static/sessions.js b/static/sessions.js
index 4072f42c..51a9e7b4 100644
--- a/static/sessions.js
+++ b/static/sessions.js
@@ -276,8 +276,22 @@ async function newSession(flash){
   // default_model (from Settings) takes priority over the chat-header dropdown
   // value, which reflects the *previous* session's model. Fall back to the
   // dropdown value only when no default_model is configured.
-  const newModel=window._defaultModel||$('modelSelect').value;
-  const data=await api('/api/session/new',{method:'POST',body:JSON.stringify({model:newModel,workspace:inheritWs,profile:S.activeProfile||'default'})});
+  const modelSel=$('modelSelect');
+  const selectedDefaultModel=window._defaultModel||(modelSel&&modelSel.value)||'';
+  let defaultApplied=false;
+  if(window._defaultModel&&modelSel&&typeof _applyModelToDropdown==='function'){
+    defaultApplied=!!_applyModelToDropdown(window._defaultModel,modelSel,window._activeProvider||null);
+  }
+  const canQualify=!window._defaultModel||defaultApplied||(modelSel&&modelSel.value===selectedDefaultModel);
+  const newModelState=(canQualify&&typeof _modelStateForSelect==='function')
+    ? _modelStateForSelect(modelSel,selectedDefaultModel)
+    : {model:selectedDefaultModel,model_provider:null};
+  const data=await api('/api/session/new',{method:'POST',body:JSON.stringify({
+    model:newModelState.model,
+    model_provider:newModelState.model_provider||null,
+    workspace:inheritWs,
+    profile:S.activeProfile||'default',
+  })});
   S.session=data.session;S.messages=data.session.messages||[];
   S.lastUsage={...(data.session.last_usage||{})};
   if(flash)S.session._flash=true;
@@ -287,7 +301,7 @@ async function newSession(flash){
   // Sync chat-header dropdown to the session's model so the UI reflects
   // the default model the server actually used (#872).
   if(S.session.model && S.session.model!==$('modelSelect').value && typeof _applyModelToDropdown==='function'){
-    _applyModelToDropdown(S.session.model,$('modelSelect'));
+    _applyModelToDropdown(S.session.model,$('modelSelect'),S.session.model_provider||null);
     if(typeof syncModelChip==='function') syncModelChip();
   }
   // Reset per-session visual state: a fresh chat is idle even if another
@@ -515,8 +529,10 @@ function _resolveSessionModelForDisplaySoon(sid){
     try{
       const data=await api(`/api/session?session_id=${encodeURIComponent(sid)}&messages=0&resolve_model=1`);
       const model=data&&data.session&&data.session.model;
+      const provider=data&&data.session&&data.session.model_provider;
       if(!model||!S.session||S.session.session_id!==sid) return;
       S.session.model=model;
+      S.session.model_provider=provider||null;
       S.session._modelResolutionDeferred=false;
       syncTopbar();
     }catch(_){
@@ -897,7 +913,7 @@ function _openSessionActionMenu(session, anchorEl){
     async()=>{
       closeSessionActionMenu();
       try{
-        const res=await api('/api/session/new',{method:'POST',body:JSON.stringify({workspace:session.workspace,model:session.model})});
+        const res=await api('/api/session/new',{method:'POST',body:JSON.stringify({workspace:session.workspace,model:session.model,model_provider:session.model_provider||null})});
         if(res.session){
           await api('/api/session/rename',{method:'POST',body:JSON.stringify({session_id:res.session.session_id,title:(session.title||'Untitled')+' (copy)'})});
           await loadSession(res.session.session_id);
@@ -1248,8 +1264,13 @@ function _sessionTimeBucketLabel(timestampMs, nowMs) {
   return t('session_time_bucket_older');
 }
 
-function _sessionLineageKey(s){
+function _sessionLineageKey(s, sessionIdsInList){
   if(!s||!s.session_id) return null;
+  // If parent_session_id points to another session in the current list,
+  // this is a subagent child — don't collapse it into lineage (#494).
+  if(s.parent_session_id && sessionIdsInList && sessionIdsInList.has(s.parent_session_id)){
+    return null;
+  }
   return s._lineage_root_id || s.lineage_root_id || s.parent_session_id || null;
 }
 
@@ -1262,9 +1283,10 @@ function _sessionLineageContainsSession(s, sid){
 
 function _collapseSessionLineageForSidebar(sessions){
   const result=[];
+  const sessionIdsInList=new Set((sessions||[]).map(s=>s.session_id));
   const groups=new Map();
   for(const s of sessions||[]){
-    const key=_sessionLineageKey(s);
+    const key=_sessionLineageKey(s, sessionIdsInList);
     if(!key){result.push(s);continue;}
     if(!groups.has(key)) groups.set(key,[]);
     groups.get(key).push(s);
@@ -1308,6 +1330,23 @@ function renderSessionListFromCache(){
   // Filter archived unless toggle is on
   const sessionsRaw=_showArchived?projectFiltered:projectFiltered.filter(s=>!s.archived);
   const sessions=_collapseSessionLineageForSidebar(sessionsRaw);
+  // Build parent→children map for subagent tree (#494).
+  // Only children whose parent exists in the current (post-collapse) list are grouped.
+  const _sessionIdsInList=new Set(sessions.map(s=>s.session_id));
+  const _parentChildrenMap=new Map();
+  const _topLevelSessions=[];
+  for(const s of sessions){
+    if(s.parent_session_id && _sessionIdsInList.has(s.parent_session_id)){
+      if(!_parentChildrenMap.has(s.parent_session_id)) _parentChildrenMap.set(s.parent_session_id,[]);
+      _parentChildrenMap.get(s.parent_session_id).push(s);
+    } else {
+      _topLevelSessions.push(s);
+    }
+  }
+  // Collapse state for subagent tree groups — persisted in localStorage (#494)
+  let _treeCollapsed={};
+  try{_treeCollapsed=JSON.parse(localStorage.getItem('hermes-tree-collapsed')||'{}');}catch(e){}
+  const _saveTreeCollapsed=()=>{try{localStorage.setItem('hermes-tree-collapsed',JSON.stringify(_treeCollapsed));}catch(e){}};
   const archivedCount=projectFiltered.filter(s=>s.archived).length;
   const list=$('sessionList');list.innerHTML='';
   // Batch select bar (when in select mode)
@@ -1399,7 +1438,7 @@ function renderSessionListFromCache(){
     empty.textContent='No sessions in this project yet.';
     list.appendChild(empty);
   }
-  const orderedSessions=[...sessions].sort((a,b)=>_sessionTimestampMs(b)-_sessionTimestampMs(a));
+  const orderedSessions=[..._topLevelSessions].sort((a,b)=>_sessionTimestampMs(b)-_sessionTimestampMs(a));
   // Separate pinned from unpinned
   const pinned=orderedSessions.filter(s=>s.pinned);
   const unpinned=orderedSessions.filter(s=>!s.pinned);
@@ -1445,7 +1484,47 @@ function renderSessionListFromCache(){
       _saveCollapsed();
     };
     wrapper.appendChild(hdr);
-    for(const s of g.items){ body.appendChild(_renderOneSession(s, Boolean(g.isPinned))); }
+    for(const s of g.items){
+      const parentEl=_renderOneSession(s, Boolean(g.isPinned));
+      body.appendChild(parentEl);
+      // Render subagent children as indented tree (#494)
+      const children=_parentChildrenMap.get(s.session_id);
+      if(children&&children.length){
+        parentEl.classList.add('session-parent');
+        const treeCaret=document.createElement('span');
+        treeCaret.className='session-tree-caret';
+        treeCaret.textContent='\u25B8'; // right-pointing triangle (collapsed)
+        treeCaret.title=t('subagent_children');
+        parentEl.querySelector('.session-title-row').prepend(treeCaret);
+        const childCount=children.length;
+        const childBadge=document.createElement('span');
+        childBadge.className='session-tree-badge';
+        childBadge.textContent=childCount;
+        childBadge.title=t('subagent_children');
+        parentEl.querySelector('.session-title-row').appendChild(childBadge);
+        const isCollapsed=_treeCollapsed[s.session_id]!==false; // collapsed by default
+        const childContainer=document.createElement('div');
+        childContainer.className='session-tree-children';
+        if(isCollapsed){childContainer.style.display='none';treeCaret.classList.add('collapsed');}
+        else{treeCaret.classList.remove('collapsed');treeCaret.textContent='\u25BE';}
+        const sortedChildren=[...children].sort((a,b)=>_sessionTimestampMs(b)-_sessionTimestampMs(a));
+        for(const child of sortedChildren){
+          const childEl=_renderOneSession(child, Boolean(g.isPinned));
+          childEl.classList.add('session-tree-child');
+          childContainer.appendChild(childEl);
+        }
+        body.appendChild(childContainer);
+        treeCaret.onclick=(e)=>{
+          e.stopPropagation();
+          const hidden=childContainer.style.display==='none';
+          childContainer.style.display=hidden?'':'none';
+          treeCaret.textContent=hidden?'\u25BE':'\u25B8';
+          treeCaret.classList.toggle('collapsed',!hidden);
+          _treeCollapsed[s.session_id]=!hidden;
+          _saveTreeCollapsed();
+        };
+      }
+    }
     wrapper.appendChild(body);
     list.appendChild(wrapper);
   }
diff --git a/static/style.css b/static/style.css
index 65d5c531..b6f37d43 100644
--- a/static/style.css
+++ b/static/style.css
@@ -958,6 +958,18 @@
   @keyframes mic-pulse{0%,100%{box-shadow:0 0 0 0 rgba(239,83,80,.3);}50%{box-shadow:0 0 0 6px rgba(239,83,80,0);}}
   .mic-status{font-size:11px;color:var(--error);padding:4px 12px;display:flex;align-items:center;gap:6px;}
   .mic-dot{width:6px;height:6px;border-radius:50%;background:var(--error);animation:mic-pulse 1.2s ease-in-out infinite;flex-shrink:0;}
+
+  /* ── Turn-based voice mode (#1333) ── */
+  .voice-mode-btn{transition:color .15s,background .15s;}
+  .voice-mode-btn.active{color:var(--accent);background:rgba(var(--accent-rgb,99,102,241),.15);}
+  .voice-mode-btn.active svg{filter:drop-shadow(0 0 3px rgba(var(--accent-rgb,99,102,241),.5));}
+  .voice-mode-bar{font-size:11px;padding:4px 12px;display:flex;align-items:center;gap:8px;border-bottom:1px solid rgba(255,255,255,.05);}
+  .voice-mode-indicator{width:8px;height:8px;border-radius:50%;flex-shrink:0;}
+  .voice-mode-indicator.listening{background:var(--error);animation:voice-mode-pulse 1s ease-in-out infinite;}
+  .voice-mode-indicator.speaking{background:var(--accent);animation:voice-mode-pulse 1.5s ease-in-out infinite;}
+  .voice-mode-indicator.thinking{background:var(--warning,#f59e0b);animation:voice-mode-pulse 2s ease-in-out infinite;}
+  .voice-mode-label{color:var(--muted);font-size:11px;}
+  @keyframes voice-mode-pulse{0%,100%{opacity:1;transform:scale(1);}50%{opacity:.5;transform:scale(.85);}}
   .status-text{font-size:11px;color:var(--muted);padding-left:4px;}
   .send-btn{width:34px;height:34px;border-radius:50%;background:var(--accent);border:none;color:#fff;cursor:pointer;display:flex;align-items:center;justify-content:center;flex-shrink:0;transition:background .15s,transform .15s,box-shadow .15s;box-shadow:0 2px 8px var(--accent-bg-strong);}
   .send-btn.stop,.send-btn.interrupt{background:var(--error);box-shadow:0 2px 10px rgba(0,0,0,.18);}
@@ -1253,7 +1265,7 @@
     #terminalWorkspaceLabel{max-width:110px;}
     #terminalDockWorkspaceLabel{max-width:96px;}
     /* Touch targets — minimum 44px */
-    .icon-btn,.mic-btn{min-width:44px;min-height:44px;}
+    .icon-btn,.mic-btn,.voice-mode-btn{min-width:44px;min-height:44px;}
     .session-item{min-height:44px;padding:10px 40px 10px 12px;}
     .session-item.streaming,.session-item.unread{padding-right:40px;}
     .session-actions{opacity:1;pointer-events:auto;}
@@ -1411,8 +1423,12 @@
 .msg-row:hover .msg-actions{opacity:1;}
 .msg-action-btn{background:none;border:none;color:var(--muted);cursor:pointer;font-size:13px;padding:2px 5px;border-radius:5px;transition:color .12s,background .12s;line-height:1;}
 .msg-action-btn:hover{color:var(--accent-text);background:var(--accent-bg);}
-/* TTS speaker button: hidden by default, shown when TTS enabled */
+/* TTS speaker button: hidden by default, shown when TTS is enabled.
+ * Use body-class selector instead of JS inline-style so the rule survives
+ * subsequent renderMd() passes and is not subject to inline-style cascade
+ * collisions with the .msg-action-btn parent (#1409). */
 .msg-tts-btn{display:none;}
+body.tts-enabled .msg-tts-btn{display:inline-flex;align-items:center;}
 .msg-tts-btn[data-speaking="1"]{color:var(--accent);animation:tts-pulse 1s ease-in-out infinite;}
 @keyframes tts-pulse{0%,100%{opacity:1}50%{opacity:.5}}
 
@@ -1996,8 +2012,9 @@ main.main > #mainSkills,
 main.main > #mainMemory,
 main.main > #mainTasks,
 main.main > #mainWorkspaces,
-main.main > #mainProfiles{display:none;}
-main.main:not(.showing-settings):not(.showing-skills):not(.showing-memory):not(.showing-tasks):not(.showing-workspaces):not(.showing-profiles) > #mainChat{display:flex;}
+main.main > #mainProfiles,
+main.main > #mainInsights{display:none;}
+main.main:not(.showing-settings):not(.showing-skills):not(.showing-memory):not(.showing-tasks):not(.showing-workspaces):not(.showing-profiles):not(.showing-insights) > #mainChat{display:flex;}
 main.main.showing-settings > #mainSettings{display:flex;overflow-y:auto;}
 main.main.showing-skills > #mainSkills{display:flex;}
 main.main.showing-memory > #mainMemory{display:flex;}
@@ -2298,6 +2315,17 @@ main.main.showing-profiles > #mainProfiles{display:flex;}
 .session-item.archived{opacity:.5;}
 .session-item.archived .session-title{font-style:italic;}
 
+/* ── Subagent session tree (#494) ── */
+.session-tree-children{margin-left:16px;border-left:1px solid var(--border,rgba(255,255,255,.1));padding-left:4px;}
+.session-tree-child.session-item{font-size:12px;opacity:.85;border-radius:6px;padding:6px 8px;}
+.session-tree-child.session-item:hover{opacity:1;}
+.session-tree-child.session-item.active{opacity:1;}
+.session-tree-child.session-item .session-meta{font-size:10px;}
+.session-tree-caret{display:inline-flex;align-items:center;justify-content:center;width:16px;height:16px;font-size:10px;cursor:pointer;color:var(--muted);flex-shrink:0;transition:transform .15s;user-select:none;border-radius:3px;margin-right:2px;}
+.session-tree-caret:hover{color:var(--text);background:rgba(255,255,255,.06);}
+.session-tree-caret.collapsed{transform:none;}
+.session-tree-badge{display:inline-flex;align-items:center;justify-content:center;min-width:16px;height:16px;font-size:9px;font-weight:700;padding:0 4px;border-radius:8px;background:rgba(99,179,237,.2);color:#63b3ed;margin-left:auto;flex-shrink:0;user-select:none;}
+
 /* ── Session tags ── */
 .session-tag{display:inline-block;font-size:9px;font-weight:600;padding:1px 5px;margin-left:4px;border-radius:3px;background:rgba(99,179,237,.2);color:#63b3ed;cursor:pointer;vertical-align:middle;}
 .session-tag:hover{background:rgba(99,179,237,.35);}
@@ -2856,3 +2884,40 @@ main.main > .main-view:not([id="mainChat"]):not([id="mainSettings"]) .main-view-
 .html-preview-iframe{width:100%;height:400px;border:none;display:block;background:#fff;}
 .html-preview-fallback{padding:8px;font-size:13px;}
 .html-preview-spinner{animation:pulse 1.5s ease-in-out infinite;}
+
+/* ── Insights panel (#464) ────────────────────────────────────────────────── */
+main.main.showing-insights > #mainInsights{display:flex;overflow-y:auto;}
+.insights-grid{display:grid;grid-template-columns:repeat(auto-fit,minmax(140px,1fr));gap:12px;margin-bottom:16px;}
+.insights-stat{background:var(--surface-2);border:1px solid var(--border);border-radius:8px;padding:14px;}
+.insights-stat-value{font-size:22px;font-weight:700;color:var(--text);}
+.insights-stat-label{font-size:11px;color:var(--muted);margin-top:4px;}
+.insights-row{display:grid;grid-template-columns:1fr 1fr;gap:12px;margin-bottom:16px;}
+.insights-card{background:var(--surface-2);border:1px solid var(--border);border-radius:8px;padding:14px;}
+.insights-card-title{font-size:13px;font-weight:600;color:var(--text);margin-bottom:10px;}
+.insights-table{width:100%;font-size:12px;}
+.insights-table-head{display:grid;grid-template-columns:1fr 80px;padding:4px 0;border-bottom:1px solid var(--border);font-weight:600;color:var(--muted);font-size:11px;}
+.insights-table-row{display:grid;grid-template-columns:1fr 80px;padding:6px 0;border-bottom:1px solid var(--border,.05);}
+.insights-model-name{overflow:hidden;text-overflow:ellipsis;white-space:nowrap;}
+.insights-bars{display:flex;flex-direction:column;gap:6px;}
+.insights-bar-row{display:grid;grid-template-columns:40px 1fr 40px;align-items:center;gap:8px;}
+.insights-bar-label{font-size:11px;color:var(--muted);text-align:right;}
+.insights-bar-track{height:16px;background:var(--border,.15);border-radius:4px;overflow:hidden;}
+.insights-bar-fill{height:100%;background:var(--accent);border-radius:4px;min-width:2px;transition:width .3s;}
+.insights-bar-fill.peak{background:#f6ad55;}
+.insights-bar-value{font-size:11px;color:var(--text);}
+.insights-token-row{display:flex;justify-content:space-between;padding:4px 0;font-size:12px;border-bottom:1px solid var(--border,.05);}
+.insights-token-label{color:var(--muted);}
+.insights-token-value{font-weight:600;}
+
+/* ── Checkpoints / Rollback UI (#466) ─────────────────────────────────────── */
+.checkpoint-list{display:flex;flex-direction:column;gap:8px;}
+.checkpoint-item{display:flex;align-items:center;justify-content:space-between;padding:8px 10px;background:var(--surface-2);border:1px solid var(--border);border-radius:6px;font-size:12px;}
+.checkpoint-item-actions{display:flex;gap:6px;}
+.checkpoint-item-actions button{background:none;border:none;color:var(--muted);cursor:pointer;padding:2px 4px;border-radius:4px;}
+.checkpoint-item-actions button:hover{color:var(--accent);background:rgba(255,255,255,.06);}
+.checkpoint-item-actions button.danger:hover{color:#fc8181;}
+.checkpoint-diff{position:fixed;inset:0;z-index:1000;background:rgba(0,0,0,.6);display:flex;align-items:center;justify-content:center;}
+.checkpoint-diff-modal{background:var(--surface);border:1px solid var(--border);border-radius:8px;max-width:700px;width:90%;max-height:80vh;display:flex;flex-direction:column;}
+.checkpoint-diff-header{display:flex;justify-content:space-between;align-items:center;padding:12px 16px;border-bottom:1px solid var(--border);}
+.checkpoint-diff-body{padding:12px 16px;overflow-y:auto;flex:1;}
+.checkpoint-diff-body pre{font-size:11px;line-height:1.4;white-space:pre-wrap;word-break:break-all;}
diff --git a/static/ui.js b/static/ui.js
index ad0a1da3..3674c338 100644
--- a/static/ui.js
+++ b/static/ui.js
@@ -225,6 +225,7 @@ setTimeout(_initMediaPlaybackObserver,0);
 // Dynamic model labels -- populated by populateModelDropdown(), fallback to static map
 let _dynamicModelLabels={};
 window._configuredModelBadges=window._configuredModelBadges||{};
+const MODEL_STATE_KEY='hermes-webui-model-state';
 
 // ── Smart model resolver ────────────────────────────────────────────────────
 // Finds the best matching option value in a <select> for a given model ID.
@@ -239,9 +240,60 @@ function _getOptionProviderId(opt){
     return group.dataset.provider;
   }
   const value=String(opt.value||'');
-  if(value.startsWith('@') && value.includes(':')) return value.slice(1,value.indexOf(':'));
+  if(value.startsWith('@') && value.includes(':')) return value.slice(1,value.lastIndexOf(':'));
   return '';
 }
+function _providerFromModelValue(modelId){
+  const value=String(modelId||'').trim();
+  if(value.startsWith('@')&&value.includes(':')) return value.slice(1,value.lastIndexOf(':'));
+  return '';
+}
+function _modelStateForSelect(sel, modelId){
+  const value=String(modelId||'').trim();
+  if(!value) return {model:'',model_provider:null};
+  const explicitProvider=_providerFromModelValue(value);
+  if(explicitProvider) return {model:value,model_provider:explicitProvider};
+  const opt=sel&&sel.selectedOptions&&sel.selectedOptions[0];
+  const provider=String(_getOptionProviderId(opt)||'').trim();
+  return {model:value,model_provider:(provider&&provider!=='default')?provider:null};
+}
+function _providerQualifiedModelValueForSelect(sel, modelId){
+  return _modelStateForSelect(sel,modelId).model;
+}
+function _readPersistedModelState(){
+  try{
+    const raw=localStorage.getItem(MODEL_STATE_KEY);
+    if(raw){
+      const parsed=JSON.parse(raw);
+      if(parsed&&parsed.model){
+        return {
+          model:String(parsed.model||''),
+          model_provider:parsed.model_provider?String(parsed.model_provider):(_providerFromModelValue(parsed.model)||null),
+        };
+      }
+    }
+  }catch(_){}
+  const legacy=localStorage.getItem('hermes-webui-model');
+  if(!legacy) return null;
+  return {model:legacy,model_provider:_providerFromModelValue(legacy)||null};
+}
+function _writePersistedModelState(model, modelProvider){
+  const value=String(model||'').trim();
+  const provider=modelProvider?String(modelProvider).trim():(_providerFromModelValue(value)||null);
+  if(!value){
+    localStorage.removeItem('hermes-webui-model');
+    localStorage.removeItem(MODEL_STATE_KEY);
+    return;
+  }
+  localStorage.setItem('hermes-webui-model', value);
+  try{
+    localStorage.setItem(MODEL_STATE_KEY, JSON.stringify({model:value,model_provider:provider||null}));
+  }catch(_){}
+}
+function _clearPersistedModelState(){
+  localStorage.removeItem('hermes-webui-model');
+  localStorage.removeItem(MODEL_STATE_KEY);
+}
 function _findModelInDropdown(modelId, sel, preferredProviderId){
   if(!modelId||!sel) return null;
   const options=Array.from(sel.options);
@@ -250,7 +302,12 @@ function _findModelInDropdown(modelId, sel, preferredProviderId){
   // Also strip @provider: prefix from deduplicated model IDs (#1228, #1313).
   const norm=s=>s.toLowerCase().replace(/^[^/]+\//,'').replace(/^@([^:]+:)+/,'').replace(/-/g,'.');
   const target=norm(modelId);
-  const preferred=String(preferredProviderId||'').toLowerCase();
+  let explicitProvider='';
+  const rawModel=String(modelId||'');
+  if(rawModel.startsWith('@')&&rawModel.includes(':')){
+    explicitProvider=rawModel.slice(1,rawModel.lastIndexOf(':'));
+  }
+  const preferred=String(preferredProviderId||explicitProvider||'').toLowerCase();
   if(preferred){
     const providerMatch=options.find(o=>norm(o.value)===target && _getOptionProviderId(o).toLowerCase()===preferred);
     if(providerMatch) return providerMatch.value;
@@ -314,7 +371,7 @@ async function populateModelDropdown(){
       sel.appendChild(og);
     }
     // Set default model from server if no localStorage preference
-    if(data.default_model && !localStorage.getItem('hermes-webui-model')){
+    if(data.default_model && !(typeof _readPersistedModelState==='function'&&_readPersistedModelState()) && !localStorage.getItem('hermes-webui-model')){
       _applyModelToDropdown(data.default_model, sel, data.active_provider||null);
     }
     if(typeof syncModelChip==='function') syncModelChip();
@@ -367,7 +424,7 @@ function _addLiveModelsToSelect(provider, models, sel){
   const existingNorm=new Set([...sel.options].map(o=>_normId(o.value)));
   let added=0;
   const _ap=(window._activeProvider||'').toLowerCase();
-  const _isPortalFetch=_ap && _ap!=='openrouter' && _ap!=='custom' && provider===_ap;
+  const _isPortalFetch=_ap && _ap!=='openrouter' && _ap!=='custom' && _ap!=='openai-codex' && provider===_ap;
   for(const m of models){
     let mid=m.id;
     if(_isPortalFetch && !mid.startsWith('@')){
@@ -383,13 +440,14 @@ function _addLiveModelsToSelect(provider, models, sel){
     _dynamicModelLabels[mid]=m.label||m.id;
     added++;
   }
-  if(added>0 && currentVal) _applyModelToDropdown(currentVal, sel);
+  const currentProvider=(S.session&&S.session.model_provider)||null;
+  if(added>0 && currentVal) _applyModelToDropdown(currentVal, sel, currentProvider);
   // After live models are added, re-apply the session's model in case it was
   // absent from the static list and syncTopbar() fired before the live fetch
   // completed (#1169). This ensures the session model wins over any premature
   // fallback that may have set sel.value to the first available option.
   if(S.session && S.session.model && sel.id==='modelSelect'){
-    const reapplied=_applyModelToDropdown(S.session.model, sel);
+    const reapplied=_applyModelToDropdown(S.session.model, sel, S.session.model_provider||null);
     if(reapplied && typeof syncModelChip==='function') syncModelChip();
   }
   return added;
@@ -1885,7 +1943,10 @@ function setBusy(v){
         // Note: profile is NOT restored — full profile switch requires server interaction
         if(next.model&&S.session&&next.model!==S.session.model){
           S.session.model=next.model;
-          if(typeof _applyModelToDropdown==='function'&&$('modelSelect')) _applyModelToDropdown(next.model,$('modelSelect'));
+        }
+        if(next.model_provider&&S.session) S.session.model_provider=next.model_provider;
+        if(next.model&&S.session){
+          if(typeof _applyModelToDropdown==='function'&&$('modelSelect')) _applyModelToDropdown(next.model,$('modelSelect'),S.session.model_provider||null);
           if(typeof syncModelChip==='function') syncModelChip();
         }
         autoResize();
@@ -1968,8 +2029,9 @@ function _renderQueueChips(sid){
       const _doMerge=(snapshot)=>{
         const combined=snapshot.map(e=>e&&(e.text||e.message||e.content||'')).filter(Boolean).join('\n\n');
         const liveQ=_getSessionQueue(sid,false);
+        const first=snapshot.find(e=>e)||{};
         const firstFiles=(snapshot.find(e=>e&&Array.isArray(e.files)&&e.files.length)||{files:[]}).files;
-        liveQ.length=0;liveQ.push({text:combined,files:firstFiles,_queued_at:Date.now()});
+        liveQ.length=0;liveQ.push({text:combined,files:firstFiles,model:first.model||'',model_provider:first.model_provider||null,_queued_at:Date.now()});
         SESSION_QUEUES[sid]=liveQ;
         try{sessionStorage.setItem('hermes-queue-'+sid,JSON.stringify(liveQ));}catch(_){}
         delete _queueRenderKeys[sid];
@@ -2731,10 +2793,12 @@ function syncTopbar(){
   let currentModel=S.session.model||'';
   if(modelOverride){
     S._pendingProfileModel=null;
-    _applyModelToDropdown(modelOverride,$('modelSelect'));
+    const providerOverride=S._pendingProfileModelProvider||null;
+    S._pendingProfileModelProvider=null;
+    _applyModelToDropdown(modelOverride,$('modelSelect'),providerOverride);
     currentModel=modelOverride;
   } else {
-    const applied=_applyModelToDropdown(currentModel,$('modelSelect'));
+    const applied=_applyModelToDropdown(currentModel,$('modelSelect'),S.session.model_provider||null);
     // If the model isn't in the current provider list, silently reset to the
     // first available model so stale values don't pollute the picker (#829).
     if(!applied && currentModel){
@@ -2756,11 +2820,12 @@ function syncTopbar(){
           modelSel.value=first.value;
           if(!deferModelCorrection){
             S.session.model=first.value;
+            S.session.model_provider=_getOptionProviderId(first)||null;
             // Persist the correction so the session doesn't re-inject on next load.
             fetch(new URL('api/session/update',document.baseURI||location.href).href,{
               method:'POST',credentials:'include',
               headers:{'Content-Type':'application/json'},
-              body:JSON.stringify({session_id:S.session.id||S.session.session_id,model:first.value})
+              body:JSON.stringify({session_id:S.session.id||S.session.session_id,model:first.value,model_provider:S.session.model_provider||null})
             }).catch(()=>{});
           }
         }
diff --git a/tests/test_465_session_branching.py b/tests/test_465_session_branching.py
index d8a68b7f..058a0862 100644
--- a/tests/test_465_session_branching.py
+++ b/tests/test_465_session_branching.py
@@ -235,8 +235,8 @@ def test_parent_indicator_clickable():
         src = f.read()
     # Find the parent indicator block
     parent_block = re.search(
-        r'parent_session_id\).*?(?=titleRow\.appendChild)',
-        src, re.DOTALL
+        r'branch-indicator[\s\S]*?parent_session_id[\s\S]*?titleRow\.appendChild',
+        src
     )
     assert parent_block, "Could not find parent indicator block"
     block = parent_block.group(0)
diff --git a/tests/test_499_tts_playback.py b/tests/test_499_tts_playback.py
index 21630dc9..efbc1933 100644
--- a/tests/test_499_tts_playback.py
+++ b/tests/test_499_tts_playback.py
@@ -193,3 +193,57 @@ class TestTtsStyles:
         src = _read('style.css')
         assert 'tts-pulse' in src, \
             "tts-pulse animation not found in style.css"
+
+
+class TestIssue1409TtsToggleBodyClass:
+    """Regression: #1409 — TTS toggle had no effect because of CSS specificity collision.
+
+    Original bug: ``_applyTtsEnabled`` set ``btn.style.display=enabled?'':'none'``.
+    The empty-string branch removes the inline override, after which the
+    ``.msg-tts-btn { display:none; }`` rule from style.css applies — so both
+    "enabled" and "disabled" states left the button hidden.
+
+    Fix: toggle a body-level class (``body.tts-enabled``) and gate the speaker
+    icon on a compound selector ``body.tts-enabled .msg-tts-btn``. This bypasses
+    the inline-style cascade collision and survives ``renderMd()`` re-renders.
+    """
+
+    def test_apply_tts_enabled_uses_body_class(self):
+        """_applyTtsEnabled must toggle the document body's `tts-enabled` class."""
+        src = _read('panels.js')
+        # The new shape: toggle body class instead of writing inline display
+        assert "document.body.classList.toggle('tts-enabled'" in src, (
+            "_applyTtsEnabled must toggle the body.tts-enabled class — see #1409. "
+            "Reverting to inline `style.display` will silently break the toggle "
+            "again because of the .msg-action-btn / .msg-tts-btn cascade."
+        )
+
+    def test_apply_tts_enabled_does_not_use_inline_display(self):
+        """_applyTtsEnabled must NOT set inline `style.display` on .msg-tts-btn."""
+        src = _read('panels.js')
+        # Find the function body and check it doesn't set inline display
+        # on individual buttons (the broken pattern).
+        m = re.search(
+            r'function _applyTtsEnabled\([^)]*\)\s*\{(?P<body>[^}]*)\}',
+            src,
+        )
+        assert m, "_applyTtsEnabled function body not found in panels.js"
+        body = m.group('body')
+        assert '.style.display' not in body, (
+            "_applyTtsEnabled body must not set inline style.display — that's "
+            "the #1409 bug. Use body.classList.toggle('tts-enabled') instead."
+        )
+
+    def test_body_class_selector_in_css(self):
+        """style.css must show .msg-tts-btn only when body.tts-enabled is set."""
+        src = _read('style.css')
+        assert 'body.tts-enabled .msg-tts-btn' in src, (
+            "Missing `body.tts-enabled .msg-tts-btn` selector in style.css — "
+            "without this rule the body class has no visual effect (#1409)."
+        )
+        # The default-hidden rule must still be present (so no body class = no icon).
+        assert '.msg-tts-btn{display:none;}' in src or \
+               re.search(r'\.msg-tts-btn\s*\{[^}]*display\s*:\s*none', src), (
+            "Default `.msg-tts-btn{display:none;}` rule must remain so the "
+            "icon is hidden by default (#1409)."
+        )
diff --git a/tests/test_issue856_pinned_indicator_layout.py b/tests/test_issue856_pinned_indicator_layout.py
index aeb0f412..f90cc7b0 100644
--- a/tests/test_issue856_pinned_indicator_layout.py
+++ b/tests/test_issue856_pinned_indicator_layout.py
@@ -11,7 +11,8 @@ def test_pinned_indicator_renders_inside_title_row():
     title_row_idx = SESSIONS_JS.find("titleRow.className='session-title-row';")
     assert title_row_idx != -1, "session title row construction not found"
 
-    assert "body.appendChild(_renderOneSession(s, Boolean(g.isPinned)))" in SESSIONS_JS
+    assert ("body.appendChild(_renderOneSession(s, Boolean(g.isPinned)))" in SESSIONS_JS
+            or "body.appendChild(parentEl)" in SESSIONS_JS)
     assert "function _renderOneSession(s, isPinnedGroup=false)" in SESSIONS_JS
     assert "if(s.pinned&&!isPinnedGroup){" in SESSIONS_JS
 
diff --git a/tests/test_model_picker_badges.py b/tests/test_model_picker_badges.py
index 3b2ffa73..b14ec5c8 100644
--- a/tests/test_model_picker_badges.py
+++ b/tests/test_model_picker_badges.py
@@ -97,6 +97,19 @@ def test_ui_badge_lookup_prefers_row_provider_for_duplicate_model_ids():
     assert "const providerMatch=matches.find(badge=>String(badge&&badge.provider||'').toLowerCase()===provider);" in js
 
 
+def test_configured_model_group_label_has_i18n_key():
+    """The Configured model group must not render the raw i18n key."""
+    root = Path(__file__).resolve().parent.parent
+    i18n = (root / "static" / "i18n.js").read_text(encoding="utf-8")
+
+    locale_count = i18n.count("_lang:")
+    key_count = i18n.count("model_group_configured:")
+    assert key_count == locale_count, (
+        "model_group_configured must be present in every locale block so "
+        "t('model_group_configured') never falls back to the raw key."
+    )
+
+
 def test_get_available_models_cache_preserves_configured_model_badges(tmp_path, monkeypatch):
     cache_path = tmp_path / "models_cache.json"
     old_cfg = config.cfg
diff --git a/tests/test_provider_management.py b/tests/test_provider_management.py
index 05adb75c..002f6ade 100644
--- a/tests/test_provider_management.py
+++ b/tests/test_provider_management.py
@@ -372,3 +372,92 @@ class TestProvidersEndpoints:
         """POST /api/providers/delete without provider should return 400."""
         body, status = _post("/api/providers/delete", {})
         assert status == 400
+
+
+class TestIssue1410OllamaEnvVarBleed:
+    """Regression: Ollama Cloud key must not flip local Ollama to has_key=True.
+
+    Both providers used to share OLLAMA_API_KEY in _PROVIDER_ENV_VAR. After
+    a user added a key for Ollama Cloud, the local Ollama card also lit up
+    "API key configured" — incorrect because the runtime in
+    hermes_cli/runtime_provider.py only consumes OLLAMA_API_KEY when the
+    base URL hostname is ollama.com. Local Ollama is keyless by default.
+
+    Fix: drop bare "ollama" from _PROVIDER_ENV_VAR so the env-var check is
+    only applied to ollama-cloud. Local Ollama users who genuinely need a
+    key can still set providers.ollama.api_key in config.yaml.
+    """
+
+    def test_ollama_local_not_configured_when_only_cloud_env_var_set(
+        self, monkeypatch, tmp_path,
+    ):
+        """OLLAMA_API_KEY in env should mark ollama-cloud configured but not bare ollama."""
+        _install_fake_hermes_cli(monkeypatch)
+        monkeypatch.setattr(profiles, "get_active_hermes_home", lambda: tmp_path)
+        monkeypatch.setenv("OLLAMA_API_KEY", "sk-cloud-key-xyz")
+
+        old_cfg = dict(config.cfg)
+        old_mtime = config._cfg_mtime
+        config.cfg.clear()
+        config.cfg["model"] = {}
+        try:
+            config._cfg_mtime = config.Path(config._get_config_path()).stat().st_mtime
+        except Exception:
+            config._cfg_mtime = 0.0
+
+        from api.providers import get_providers
+        try:
+            result = get_providers()
+            by_id = {p["id"]: p for p in result["providers"]}
+            assert "ollama-cloud" in by_id, "ollama-cloud should appear in provider list"
+            assert "ollama" in by_id, "ollama (local) should appear in provider list"
+            assert by_id["ollama-cloud"]["has_key"] is True, \
+                "ollama-cloud should be has_key=True when OLLAMA_API_KEY is set"
+            assert by_id["ollama"]["has_key"] is False, (
+                "ollama (local) must NOT be has_key=True when only the cloud env "
+                "var is set — local Ollama is keyless and shares no env var with "
+                "Ollama Cloud (#1410)."
+            )
+            # ollama-cloud should be configurable, but local ollama should not
+            # (it has no env var mapping — keys go through providers.ollama.api_key
+            # in config.yaml if the user explicitly opts in).
+            assert by_id["ollama-cloud"]["configurable"] is True
+            assert by_id["ollama"]["configurable"] is False
+        finally:
+            config.cfg.clear()
+            config.cfg.update(old_cfg)
+            config._cfg_mtime = old_mtime
+
+    def test_ollama_local_still_configured_via_config_yaml(
+        self, monkeypatch, tmp_path,
+    ):
+        """providers.ollama.api_key in config.yaml should still mark local ollama configured."""
+        _install_fake_hermes_cli(monkeypatch)
+        monkeypatch.setattr(profiles, "get_active_hermes_home", lambda: tmp_path)
+        # Important: clear the env var so the only signal is config.yaml.
+        monkeypatch.delenv("OLLAMA_API_KEY", raising=False)
+
+        old_cfg = dict(config.cfg)
+        old_mtime = config._cfg_mtime
+        config.cfg.clear()
+        config.cfg["model"] = {}
+        config.cfg["providers"] = {"ollama": {"api_key": "local-token-abc"}}
+        try:
+            config._cfg_mtime = config.Path(config._get_config_path()).stat().st_mtime
+        except Exception:
+            config._cfg_mtime = 0.0
+
+        from api.providers import get_providers
+        try:
+            result = get_providers()
+            by_id = {p["id"]: p for p in result["providers"]}
+            assert by_id["ollama"]["has_key"] is True, (
+                "Local Ollama users with providers.ollama.api_key in config.yaml "
+                "should still report configured (#1410 fix must not regress this)."
+            )
+            # And ollama-cloud should NOT be configured by ollama's config entry.
+            assert by_id["ollama-cloud"]["has_key"] is False
+        finally:
+            config.cfg.clear()
+            config.cfg.update(old_cfg)
+            config._cfg_mtime = old_mtime
diff --git a/tests/test_provider_mismatch.py b/tests/test_provider_mismatch.py
index 1efe347c..66ce09b8 100644
--- a/tests/test_provider_mismatch.py
+++ b/tests/test_provider_mismatch.py
@@ -294,6 +294,274 @@ def test_api_models_includes_active_provider():
     )
 
 
+def test_codex_provider_qualified_model_routes_to_codex_not_openrouter():
+    """@openai-codex:gpt-5.5 must route through OpenAI Codex, not OpenRouter."""
+    import api.config as config
+
+    old_cfg = dict(config.cfg)
+    config.cfg["model"] = {
+        "provider": "openrouter",
+        "base_url": "https://openrouter.ai/api/v1",
+    }
+    try:
+        model, provider, base_url = config.resolve_model_provider(
+            "@openai-codex:gpt-5.5"
+        )
+    finally:
+        config.cfg.clear()
+        config.cfg.update(old_cfg)
+
+    assert model == "gpt-5.5"
+    assert provider == "openai-codex"
+    assert provider != "openrouter"
+    assert base_url is None
+
+
+def test_default_model_save_persists_codex_provider_for_qualified_model(tmp_path, monkeypatch):
+    """Saving @openai-codex:gpt-5.5 must persist model.provider=openai-codex."""
+    import yaml
+    import api.config as config
+
+    config_file = tmp_path / "config.yaml"
+    config_file.write_text(
+        "model:\n"
+        "  provider: openrouter\n"
+        "  default: openai/gpt-5.4\n"
+        "  base_url: https://openrouter.ai/api/v1\n",
+        encoding="utf-8",
+    )
+    old_cfg = dict(config.cfg)
+    old_mtime = config._cfg_mtime
+    monkeypatch.setattr(config, "_get_config_path", lambda: config_file)
+    config.cfg["model"] = {
+        "provider": "openrouter",
+        "default": "openai/gpt-5.4",
+        "base_url": "https://openrouter.ai/api/v1",
+    }
+    config._cfg_mtime = config_file.stat().st_mtime
+    try:
+        result = config.set_hermes_default_model("@openai-codex:gpt-5.5")
+        saved = yaml.safe_load(config_file.read_text(encoding="utf-8"))
+    finally:
+        config.cfg.clear()
+        config.cfg.update(old_cfg)
+        config._cfg_mtime = old_mtime
+        config.invalidate_models_cache()
+
+    assert result["ok"] is True
+    assert result["model"] == "gpt-5.5"
+    assert saved["model"]["default"] == "gpt-5.5"
+    assert saved["model"]["provider"] == "openai-codex"
+    assert saved["model"].get("base_url") != "https://openrouter.ai/api/v1"
+
+
+def test_active_codex_at_provider_session_model_preserved(monkeypatch):
+    """@openai-codex:gpt-5.5 session selections must keep their provider hint."""
+    import api.routes as routes
+
+    monkeypatch.setattr(
+        routes,
+        "get_available_models",
+        lambda: {
+            "active_provider": "openai-codex",
+            "default_model": "gpt-5.5",
+            "groups": [
+                {
+                    "provider": "OpenAI Codex",
+                    "provider_id": "openai-codex",
+                    "models": [{"id": "gpt-5.5", "label": "GPT-5.5"}],
+                },
+                {
+                    "provider": "OpenRouter",
+                    "provider_id": "openrouter",
+                    "models": [{"id": "openai/gpt-5.5", "label": "GPT-5.5"}],
+                },
+            ],
+        },
+    )
+
+    effective, changed = routes._resolve_compatible_session_model(
+        "@openai-codex:gpt-5.5"
+    )
+
+    assert changed is False
+    assert effective == "@openai-codex:gpt-5.5"
+
+
+def test_bare_codex_gpt_session_model_gets_separate_provider_context(monkeypatch):
+    """A bare GPT model under active Codex stays bare and carries model_provider."""
+    import api.routes as routes
+
+    monkeypatch.setattr(
+        routes,
+        "get_available_models",
+        lambda: {
+            "active_provider": "openai-codex",
+            "default_model": "gpt-5.5",
+            "groups": [
+                {
+                    "provider": "OpenAI Codex",
+                    "provider_id": "openai-codex",
+                    "models": [{"id": "gpt-5.5", "label": "GPT-5.5"}],
+                },
+                {
+                    "provider": "OpenRouter",
+                    "provider_id": "openrouter",
+                    "models": [{"id": "openai/gpt-5.5", "label": "GPT-5.5"}],
+                },
+            ],
+        },
+    )
+
+    effective, provider, changed = routes._resolve_compatible_session_model_state("gpt-5.5")
+
+    assert changed is False
+    assert effective == "gpt-5.5"
+    assert provider == "openai-codex"
+
+
+def test_session_model_normalizer_keeps_bare_codex_model_and_saves_provider(monkeypatch):
+    """Write-path normalization must persist model_provider without adding @."""
+    import api.routes as routes
+
+    monkeypatch.setattr(
+        routes,
+        "get_available_models",
+        lambda: {
+            "active_provider": "openai-codex",
+            "default_model": "gpt-5.5",
+            "groups": [
+                {
+                    "provider": "OpenAI Codex",
+                    "provider_id": "openai-codex",
+                    "models": [{"id": "gpt-5.5", "label": "GPT-5.5"}],
+                },
+            ],
+        },
+    )
+
+    save_calls = []
+
+    class DummySession:
+        def __init__(self):
+            self.model = "gpt-5.5"
+            self.model_provider = None
+
+        def save(self, touch_updated_at=True):
+            save_calls.append(touch_updated_at)
+
+    session = DummySession()
+    effective = routes._normalize_session_model_in_place(session)
+
+    assert effective == "gpt-5.5"
+    assert session.model == "gpt-5.5"
+    assert session.model_provider == "openai-codex"
+    assert save_calls == [False]
+
+
+def test_bare_codex_gpt_runtime_bridge_routes_to_codex(monkeypatch):
+    """Bare model + model_provider=openai-codex must route Codex at runtime."""
+    import api.config as config
+
+    old_cfg = dict(config.cfg)
+    config.cfg["model"] = {
+        "provider": "openrouter",
+        "default": "openai/gpt-5.4",
+        "base_url": "https://openrouter.ai/api/v1",
+    }
+    try:
+        runtime_model = config.model_with_provider_context(
+            "gpt-5.5",
+            "openai-codex",
+        )
+        model, provider, base_url = config.resolve_model_provider(runtime_model)
+    finally:
+        config.cfg.clear()
+        config.cfg.update(old_cfg)
+
+    assert runtime_model == "@openai-codex:gpt-5.5"
+    assert model == "gpt-5.5"
+    assert provider == "openai-codex"
+    assert base_url is None
+
+
+def test_non_openrouter_slash_model_provider_context_stays_unqualified():
+    """Portal/custom slash IDs must not be blindly wrapped as @provider:model."""
+    import api.config as config
+
+    runtime_model = config.model_with_provider_context(
+        "anthropic/claude-sonnet-4.6",
+        "nous",
+    )
+
+    assert runtime_model == "anthropic/claude-sonnet-4.6"
+
+
+def test_api_session_new_persists_model_provider_context():
+    """POST /api/session/new returns compact session model_provider metadata."""
+    created, status = _post(
+        "/api/session/new",
+        {"model": "gpt-5.5", "model_provider": "openai-codex"},
+    )
+
+    assert status == 200
+    assert created["session"]["model"] == "gpt-5.5"
+    assert created["session"]["model_provider"] == "openai-codex"
+
+
+def test_explicit_openrouter_selection_supported_with_codex_base_url():
+    """OpenRouter slash and @openrouter selections must remain routable."""
+    import api.config as config
+
+    old_cfg = dict(config.cfg)
+    config.cfg["model"] = {
+        "provider": "openai-codex",
+        "default": "gpt-5.5",
+        "base_url": "https://chatgpt.com/backend-api/codex",
+    }
+    try:
+        slash_model, slash_provider, slash_base_url = config.resolve_model_provider(
+            "openai/gpt-5.5"
+        )
+        at_model, at_provider, at_base_url = config.resolve_model_provider(
+            "@openrouter:openai/gpt-5.5"
+        )
+    finally:
+        config.cfg.clear()
+        config.cfg.update(old_cfg)
+
+    assert slash_model == "openai/gpt-5.5"
+    assert slash_provider == "openrouter"
+    assert slash_base_url is None
+    assert at_model == "openai/gpt-5.5"
+    assert at_provider == "openrouter"
+    assert at_base_url is None
+
+
+def test_real_provider_custom_base_url_slash_model_stays_on_configured_endpoint():
+    """A real-provider proxy base_url must not be silently rerouted to OpenRouter."""
+    import api.config as config
+
+    old_cfg = dict(config.cfg)
+    config.cfg["model"] = {
+        "provider": "openai",
+        "default": "google/gemma-4-26b-a4b",
+        "base_url": "http://proxy.local/v1",
+    }
+    try:
+        model, provider, base_url = config.resolve_model_provider(
+            "google/gemma-4-26b-a4b"
+        )
+    finally:
+        config.cfg.clear()
+        config.cfg.update(old_cfg)
+
+    assert model == "gemma-4-26b-a4b"
+    assert provider == "openai"
+    assert provider != "openrouter"
+    assert base_url == "http://proxy.local/v1"
+
+
 def test_bare_gemini_session_model_normalizes_to_active_provider_default(monkeypatch):
     """Persisted bare Gemini IDs must not survive a provider switch."""
     import api.routes as routes
@@ -689,6 +957,30 @@ class TestChatStartEffectiveModelRecovery:
         assert "localStorage.setItem('hermes-webui-model', startData.effective_model)" in src, (
             "effective_model correction must update the saved model preference"
         )
+        assert "startData.effective_model_provider" in src, (
+            "send() must preserve provider context returned by /api/chat/start"
+        )
+
+
+class TestFrontendModelProviderState:
+    """Frontend model persistence should store provider separately."""
+
+    def test_boot_session_update_sends_model_provider(self):
+        src = _read("static/boot.js")
+        assert "_modelStateForSelect" in src
+        assert "model_provider:modelState.model_provider||null" in src
+
+    def test_new_session_sends_model_provider(self):
+        src = _read("static/sessions.js")
+        assert "_modelStateForSelect(modelSel,selectedDefaultModel)" in src
+        assert "model_provider:newModelState.model_provider||null" in src
+
+    def test_ui_has_json_model_state_storage(self):
+        src = _read("static/ui.js")
+        assert "hermes-webui-model-state" in src
+        assert "function _writePersistedModelState" in src
+        assert "_providerQualifiedModelValueForSelect(sel, modelId)" in src
+        assert "return _modelStateForSelect(sel,modelId).model" in src
 
 
 def test_unknown_prefix_model_passes_through_unchanged(monkeypatch):
diff --git a/tests/test_v050255_opus_followups.py b/tests/test_v050255_opus_followups.py
new file mode 100644
index 00000000..a67986ee
--- /dev/null
+++ b/tests/test_v050255_opus_followups.py
@@ -0,0 +1,220 @@
+"""Regression tests for v0.50.255 Opus pre-release follow-ups.
+
+The v0.50.255 batch (#1390 + #1405) had four Opus advisor findings:
+
+1. MUST-FIX — `api/rollback.py::checkpoint` parameter wasn't validated; the
+   path-join `_checkpoint_root() / ws_hash / checkpoint` does NOT normalize
+   `..`, so an authenticated caller could pass `../<other-ws-hash>/<sha>` and
+   read or restore from another allowlisted workspace's checkpoint store.
+   Fix: regex validation that rejects `/`, `..`, and `.`.
+
+2. SHOULD-FIX — `api/helpers.py::_redact_text` called uncached `load_settings()`
+   per string, recursed across all messages and tool_calls. For a 50-message
+   session that's hundreds of disk reads per `/api/session?session_id=X`. Fix:
+   thread `_enabled` once through `redact_session_data()`.
+
+3. SHOULD-FIX — `static/boot.js` voice mode: the patched `autoReadLastAssistant`
+   fires globally; if the user navigates to a different session between send
+   and stream completion, TTS would speak the wrong session's last assistant
+   message. Fix: capture the active session id in `_voiceModeSend` and bail
+   out in `_speakResponse` if it doesn't match.
+
+4. NIT — `api/rollback.py::_inspect_checkpoint` had a bare `Exception` in the
+   except tuple alongside specific catches, swallowing everything (incl.
+   KeyboardInterrupt's siblings). Fix: drop to the specific tuple.
+"""
+
+from __future__ import annotations
+
+import re
+from pathlib import Path
+
+import pytest
+
+REPO = Path(__file__).resolve().parents[1]
+
+
+# ── 1: rollback checkpoint id validation ─────────────────────────────────────
+
+
+def test_rollback_validates_checkpoint_id_against_path_traversal():
+    """The checkpoint param must reject `..`, `/`, and any path-component
+    traversal vector. Without this guard, a caller can join the checkpoint
+    root with `../<other-ws-hash>/<sha>` and escape the workspace allowlist
+    (Path() / '..' does NOT normalize)."""
+    src = (REPO / "api" / "rollback.py").read_text(encoding="utf-8")
+    # Validator function exists.
+    assert "def _validate_checkpoint_id(" in src, (
+        "_validate_checkpoint_id must exist as a defense-in-depth guard for "
+        "the checkpoint parameter; without it, ../<ws>/<sha> escapes the "
+        "workspace allowlist."
+    )
+    # Both diff + restore call it.
+    assert src.count("_validate_checkpoint_id(checkpoint)") >= 2, (
+        "both get_checkpoint_diff and restore_checkpoint must call "
+        "_validate_checkpoint_id() on their checkpoint parameter."
+    )
+    # Validator rejects '..' and '.' explicitly.
+    assert 'in (".", "..")' in src, (
+        "_validate_checkpoint_id must reject literal '.' and '..' explicitly "
+        "(not just rely on the regex)."
+    )
+
+
+def test_rollback_validate_checkpoint_id_runtime_behavior():
+    """End-to-end test of the validator: traversal attempts raise ValueError."""
+    import sys
+    sys.path.insert(0, str(REPO))
+    from api.rollback import _validate_checkpoint_id
+
+    # Valid SHA-style IDs pass.
+    assert _validate_checkpoint_id("abc123def456") == "abc123def456"
+    assert _validate_checkpoint_id("a1b2c3-d4e5") == "a1b2c3-d4e5"
+    assert _validate_checkpoint_id("checkpoint_2026-05-01") == "checkpoint_2026-05-01"
+
+    # Traversal attempts blocked.
+    for bad in (
+        "../escape",
+        "..",
+        ".",
+        "../../../etc/passwd",
+        "abc/def",
+        "abc def",  # space
+        "abc\x00def",  # null byte
+        "",
+        "   ",
+        ".hidden",  # leading dot → looks like dotfile escape
+        "/abs/path",
+        "x" * 65,  # too long
+    ):
+        with pytest.raises(ValueError):
+            _validate_checkpoint_id(bad)
+
+
+# ── 2: redact_session_data settings.json read-once optimization ──────────────
+
+
+def test_redact_session_data_reads_settings_once():
+    """`redact_session_data()` must read `api_redact_enabled` ONCE per call
+    and thread it through the recursive walk via the `_enabled` keyword.
+    Calling load_settings per string was a hot-path perf regression."""
+    src = (REPO / "api" / "helpers.py").read_text(encoding="utf-8")
+
+    # The function reads settings once and threads _enabled through.
+    redact_fn_idx = src.find("def redact_session_data(")
+    assert redact_fn_idx != -1, "redact_session_data missing"
+    body = src[redact_fn_idx : redact_fn_idx + 1500]
+    assert "load_settings()" in body, (
+        "redact_session_data must read load_settings() once at the top"
+    )
+    assert body.count("_enabled=_enabled") >= 3, (
+        "redact_session_data must thread _enabled through to title, "
+        "messages, and tool_calls (3 call sites)"
+    )
+
+    # _redact_text and _redact_value accept _enabled kwarg.
+    assert "def _redact_text(text: str, *, _enabled" in src
+    assert "def _redact_value(v, *, _enabled" in src
+
+
+def test_redact_session_data_threads_enabled_once_across_recursion():
+    """End-to-end: a session payload with N strings should result in 1 read
+    of api_redact_enabled, not N. We verify by counting load_settings calls
+    via monkeypatch."""
+    import sys
+    sys.path.insert(0, str(REPO))
+    from api import helpers
+
+    call_count = [0]
+    real_load_settings = helpers.__dict__.get("load_settings")
+
+    def counting_load_settings():
+        call_count[0] += 1
+        return {"api_redact_enabled": True}
+
+    # The from-import inside redact_session_data resolves at call time, so
+    # patch in api.config where it lives.
+    from api import config
+    original = config.load_settings
+    config.load_settings = counting_load_settings
+    try:
+        # Simulate a session payload with many strings
+        session = {
+            "title": "Test session",
+            "messages": [
+                {"role": "user", "content": "hello world " * 10}
+                for _ in range(20)
+            ],
+            "tool_calls": [
+                {"name": "tool", "args": {"x": "y", "z": ["a", "b", "c"]}}
+                for _ in range(10)
+            ],
+        }
+        helpers.redact_session_data(session)
+    finally:
+        config.load_settings = original
+
+    # Should be called exactly once for the entire response, not per string.
+    assert call_count[0] == 1, (
+        f"redact_session_data called load_settings() {call_count[0]} times; "
+        f"expected exactly 1 (read-once + thread-through optimization)."
+    )
+
+
+# ── 3: voice mode session-id capture ─────────────────────────────────────────
+
+
+def test_voice_mode_speakresponse_guards_against_session_switch():
+    """The `_speakResponse` callback fires from a global override of
+    `autoReadLastAssistant`. If the user navigates to a different session
+    between sending and stream completion, the callback would TTS-read the
+    new session's last assistant message instead of the one they sent to.
+    Fix: capture session_id at thinking-time, bail in _speakResponse if it
+    doesn't match the current S.session.session_id."""
+    src = (REPO / "static" / "boot.js").read_text(encoding="utf-8")
+
+    # Session-id capture state exists.
+    assert "let _voiceModeThinkingSid=" in src, (
+        "voice mode must declare _voiceModeThinkingSid to pin the active "
+        "session id at send-time"
+    )
+
+    # _voiceModeSend captures current session_id at thinking transition.
+    send_idx = src.find("function _voiceModeSend(")
+    assert send_idx != -1
+    send_body = src[send_idx : send_idx + 1200]
+    assert "_voiceModeThinkingSid=" in send_body, (
+        "_voiceModeSend must capture the current session_id at thinking-time"
+    )
+    assert "S.session.session_id" in send_body, (
+        "_voiceModeSend must read S.session.session_id"
+    )
+
+    # _speakResponse compares current sid to captured sid and bails on mismatch.
+    speak_idx = src.find("function _speakResponse(")
+    assert speak_idx != -1
+    speak_body = src[speak_idx : speak_idx + 1500]
+    assert "_voiceModeThinkingSid" in speak_body, (
+        "_speakResponse must consult _voiceModeThinkingSid"
+    )
+    assert "_startListening()" in speak_body, (
+        "_speakResponse mismatch path must drop back to listening, not silently exit"
+    )
+
+
+# ── 4: rollback _inspect_checkpoint except tuple ─────────────────────────────
+
+
+def test_rollback_inspect_checkpoint_except_no_bare_exception():
+    """The bare `Exception` in `(subprocess.TimeoutExpired, OSError, Exception)`
+    swallowed everything including KeyboardInterrupt's siblings and made the
+    specific catches redundant. Should be the specific tuple only."""
+    src = (REPO / "api" / "rollback.py").read_text(encoding="utf-8")
+    # No bare Exception in the inspect-checkpoint except tuple.
+    assert "(subprocess.TimeoutExpired, OSError, Exception)" not in src, (
+        "_inspect_checkpoint must not catch bare Exception alongside specific "
+        "catches — the bare Exception swallows everything and makes the "
+        "specific ones redundant."
+    )
+    # The specific tuple is in place.
+    assert "(subprocess.TimeoutExpired, OSError)" in src