codex-mod

A fork of every-code/main that went a bit further than planned. Personality traits, autonomous runs, browser tools, local LLM support, session resume, and a TUI with too many settings pages.

Differences · Feature matrix · Personality · TUI · Dev notes

---

What's different

About 880 commits on top of upstream, spread across ~60 crates in code-rs/. Some are straight ports from OpenAI's codex-rs, some were ported then reshaped, and some are new.

• Codex port — pulled from upstream codex-rs into code-rs
• Hybrid — ported, then changed to fit this fork
• Fork-only — doesn't exist upstream

Note

This README only covers what's different. For general install and usage, see upstream.

• Upstream remote: upstream (just-every/code)
• Baseline branch: every-code/main

---

Feature matrix

Core

Area	Source	What it does	Where to look
Tool runtime	Codex port	Router/registry/handler instead of monolithic streaming dispatch	core/src/tools/*
Parallel tools	Codex port	Batch/exclusive scheduling; degrades when ordering metadata is missing	core/src/tools/scheduler.rs
Streaming recovery	Codex port	Retry and auto-compact reconciliation with tool-call correlation	core/src/codex/streaming/*
App-server exec v2	Hybrid	Streaming output, stdin, resize/terminate, connection-scoped PIDs	app-server/, app-server-protocol/
Windows sandbox	Hybrid	Restricted-token and elevated modes through a real sandbox backend	windows-sandbox-rs/, core/src/sandboxing.rs
MCP dispatch	Fork-only	Configurable dispatch=parallel with server/tool scheduling edits	core/src/mcp_connection_manager.rs
Network mediation	Hybrid	Temporary approvals, macOS fail-closed path, compile-gated	network-proxy/, core/src/network_approval.rs

AI and models

Area	Source	What it does	Where to look
Personality traits	Fork-only	7 trait sliders (1–5), archetype presets, tone axis — details below	core/src/personality_traits.rs, tui/.../personality/
Model families	Hybrid	Per-model context windows, truncation policy, reasoning effort clamping	core/src/model_family.rs, core/src/reasoning.rs
Auto-drive	Fork-only	Autonomous agent with Launch→Run→Resolve phases, backoff, metrics	code-auto-drive-core/
Ollama	Fork-only	Local LLM support — auto-pulls models, detects context window	ollama/
ChatGPT connector	Fork-only	Backend API for apply commands, task submission	chatgpt/
Feature flags	Fork-only	Flag registry with lifecycle stages and an experimental menu	features/
Cloud tasks	Fork-only	Task tracking UI, review coordination, diff preview	cloud-tasks/
Memories	Hybrid	Epoch-based, SQLite-indexed, published as immutable snapshots	core/src/memories/*, memories-state/

REPL and tools

Area	Source	What it does	Where to look
Multi-runtime REPL	Hybrid	Node, Deno, Python — per-runtime toggles, permission management, codex.emitImage()	core/src/tools/repl/
REPL history linkage	Fork-only	Child execs track their parent call; you can jump between them	tui/src/history_cell/
Browser automation	Fork-only	Headless Chrome via CDP — screenshots, viewport config, asset storage	browser/
search_tool_bm25	Hybrid	MCP tools stay hidden until a selection exists	core/src/tools/handlers/search_tool_bm25.rs
apply_patch	Hybrid	Wired through local safety, hook, and diff checks	core/src/tools/handlers/apply_patch.rs

TUI

Area	Source	What it does	Where to look
30 settings pages	Fork-only	Everything from model selection to shell profiles — listed below	tui/src/bottom_pane/settings_pages/
Intro animation	Fork-only	Glitch reveal, gradient effects, 10+ variants	tui/src/glitch_animation.rs
Status line	Fork-only	24 items across top/bottom lanes, clickable, configurable via /statusline	tui/src/chatwidget/status_line_flow.rs
Icon tiers	Fork-only	Nerd Fonts → Unicode → ASCII fallback, per-icon overrides in config	tui/src/icons.rs
Session resume/fork	Fork-only	JSONL session catalog — resume or fork any previous conversation	core/src/session_catalog.rs
Rate limit charts	Fork-only	Weekly/hourly usage gauges and breakdown	tui/src/rate_limits_view.rs
Hotkeys	Fork-only	Per-platform overrides (macOS, Windows, Linux, Android, Termux, BSD)	core/src/config_types.rs
Shell profiles	Fork-only	Scoped skills, MCP include/exclude, safety overrides	core/src/config/sources.rs
Auth	Fork-only	Multi-account with device code + PKCE, file/keyring/ephemeral storage	login/, core/src/auth/
Android/Termux	Fork-only	Browser/CDP paths compiled out; separate build flow	build.zsh

---

Personality system

Pick an archetype, a tone, then dial in traits. Or skip the archetype and just use the sliders.

Archetypes: Pragmatic (default) · Friendly · Concise · Enthusiastic · Mentor · None

Tone: Neutral · Formal · Casual · Direct · Encouraging

Traits (1–5 scale, 3 = neutral — neutral traits emit no instructions):

Trait	1	5
Conciseness	Very detailed	Extremely terse
Thoroughness	Trust and ship	Triple-check everything
Autonomy	Always ask first	Act without asking
Pedagogy	Just answers	Deep explanations
Enthusiasm	Reserved	High energy
Formality	Very casual	Very formal
Boldness	Conservative	Bold refactoring

The instructions sent to the model adapt based on what model you're using — reasoning models don't get told to "think step by step" (they already do), and smaller models get shorter instructions to avoid wasting context.

Edit in the TUI (Personality settings page) or in config:

[personality]
personality = "friendly"
tone = "casual"

[personality.traits]
conciseness = 4
thoroughness = 2
boldness = 5

---

The TUI

Settings pages

30 pages, accessible via shortcuts or /settings:

Category	Pages
AI	Personality, Model, Planning, Auto-drive, Agents, Prompts
Tools	REPL, MCP, Browser, Exec limits, Skills
Environment	Shell, Shell profiles, Shell escalation, Network, Secrets
Session	Memories, Review, Validation, Notifications, Experimental
Interface	Hotkeys, Status line, Theme, Verbosity, Updates, Accounts, Plugins, Apps, Overview

Icons

Three tiers with automatic fallback. Set tui.icon_mode to nerd_fonts, unicode (default), or ascii. Override individual icons if you want:

[tui.icons.gutter_exec]
ascii = "$"
unicode = "❯"
nerd_fonts = ""

Animations

The intro screen has a glitch-reveal effect with character-by-character rendering, shine bands, and fade transitions. There are 10+ reveal variants (GlitchSweep, ChromaticScan, AuroraBridge, NeonRoad, etc.) plus a header wave, shimmer effects, and a JSON-driven spinner registry.

Status line

Top and bottom lanes, each configurable with items like model name, git branch, context remaining, rate limits, REPL status, and more. Use /statusline to configure.

---

Dev notes

Rust sources live in code-rs/ (~60 crates). codex-rs/ is a read-only mirror of upstream OpenAI Codex, kept around for reference.

Build

./build-fast.sh                                       # full validation, required before push

cd code-rs && cargo check -p code-tui                 # quick incremental check
cd code-rs && cargo clippy --workspace --all-targets  # all warnings must be clean
cd code-rs && cargo nextest run --no-fail-fast         # tests

Building without the network proxy

cargo build -p code-cli                              # proxy enabled (default)
cargo build -p code-cli --no-default-features         # proxy compiled out

Compare with upstream

git diff every-code/main..main
git log --oneline every-code/main..main

Validate

./build-fast.sh