Add language indicator to Hermes-Wiki project card
Clean display-only change. Chinese / 中文 label on the Hermes-Wiki card accurately reflects the content is Chinese-only. No layout risk.
- index.html: star badge 100k+ → 105k+ (NousResearch/hermes-agent at 105k)
- models/index.html: bump meta description year 2025 → 2026
- models/index.html: update date to April 20, 2026
- models/index.html: Opus 4.6 → Opus 4.7 across all 4 sections (Overall, Coding, Writing, Research)
- New stats: 70% CursorBench, 98.5% XBOW visual-acuity, 3.75MP image resolution
- Picker card 'Complex coding' updated to Opus 4.7
- Anthropic setup box: add claude-opus-4-7 as first model name
- compare/claude-code.html: update Opus 4.6 reference to 4.7
- compare/perplexity-computer.html: update Opus 4.6 reference to 4.7
On mobile (≤640px), .nav-back span{display:none} hid both spans inside
the back link, leaving an empty bordered pill shape visible next to the
logo — the 'broken little thing' users reported seeing to the right of ⚡ Hermes.
Fix: hide .nav-back itself. The hamburger menu provides the back navigation
link, so no information is lost.
Applies to all 13 sub-pages: memory, models, eli5, community, and all 9 compare pages.
1. OpenViking link: OpenViking/OpenViking (404) → volcengine/OpenViking (22.4K stars, verified)
2. OpenViking L0/L1/L2 description: rewritten as resolution depth (50/500/full tokens),
not temperature hierarchy — this is the actual mechanism behind the token savings
3. Hindsight benchmark: standardized to 91.4–94.6% everywhere (was mixed 91–94% / 91–94.6%)
4. Hindsight link: provider name now points to hindsight.vectorize.io (product page);
GitHub link kept in meta line for developers
5. ByteRover CLI license: noted as custom license (campfirein/byterover-cli uses 'Other'),
not MIT as was implied
6. Mobile column visibility: Paid from now stays visible on mobile; Tools column hidden
instead (Stars already hidden) — pricing is more decision-relevant than tool count
7. OG/Twitter meta tags added: og:title, og:description, og:image, twitter:card,
twitter:title, twitter:description, twitter:image — shares will preview correctly
8. Holographic fact_store: card now notes fact_store exposes 9 actions
(add, search, probe, related, reason, contradict, update, remove, list)
Factual fixes:
- Hindsight link now points to github.com/vectorize-io/hindsight (hindsight.ai 403)
- Holographic: removed external link (domain for sale), note it is built into Hermes
- RetainDB: flagged domain availability as unverified
- Hindsight benchmarks added: LongMemEval 91-94.6%, BEAM 64.1%, LoCoMo 89.6%
- Holographic hosting corrected to Local only (was Cloud)
- Honcho hosting corrected to Cloud + self-host (was Cloud only)
- ByteRover license kept as Partial OSS (not MIT)
- Supermemory stars footnoted: consumer frontend, not core engine
Content additions:
- Architecture notes for all 8 providers (TEMPR, HRR/SQLite, L0/L1/L2, triple-store, etc.)
- Full pricing detail: paid tiers, per-token cloud rates
- GitHub star counts in table and provider cards
- Tool counts (2-5) in table and provider cards
- Config snippets section with 6 provider examples
- Last verified April 2026 timestamp in hero
Recommendation fixes:
- Coding agents: Hindsight #1 (TEMPR BM25, entity extraction), ByteRover #2
- Knowledge wiki: Hindsight #1 (highest LongMemEval), Supermemory #2
- Privacy: Holographic #1 (zero deps, local SQLite), then Hindsight, OpenViking
- Added Personal AI companion use case (Honcho #1)
- Benchmark winner updated to Hindsight on LongMemEval axis
Table structure:
- Best for column moved to position 2 (was last/rightmost)
- Paid from column added
- Stars and Tools columns added (hidden on narrow screens)
- Benchmark column shows multiple scores for Hindsight/ByteRover
- New page at memory/ comparing 8 providers: Mem0, Supermemory, ByteRover,
Hindsight, Holographic, OpenViking, Honcho, RetainDB
- Quick reference table: free tier, hosting, license, benchmark, best for
- Provider cards with tags, descriptions, and benchmark callouts
- Use-case picker: coding agents, knowledge wiki, all-around, free/OSS,
privacy/on-prem, enterprise — each with ranked picks + rationale
- Tradeoff snapshot: best-in-class by axis (benchmark, free tier, SDK, etc.)
- Links to official memory provider docs in CTA
- Added Memory nav link to index.html desktop nav and mobile menu
- Follows full site design system: glass nav, dark/light toggle, hamburger
- Updated to Grok 4.20: / per 1M (was /5), 2M context (was 256K)
- Arena Elo updated to 1496 (#3 overall, was 1486)
- Added 2M context pill, updated score display to Arena Elo
- history.pushState() in the smooth-scroll handler so clicking
any nav link (desktop or mobile menu) updates the URL hash.
Links like #webui, #install, #compare, etc. are now shareable.
- On-load deep-link handler: if the page is loaded with a hash
already in the URL, scroll to that section after 100ms so the
nav height is settled before calculating offset.
Coding section:
- Gemini 3.1 Pro: add Terminal-Bench 78.4% (highest of any frontier model on CLI/DevOps)
Score badge updated to show Terminal-Bench rather than SWE-bench Verified
- GPT-5.4: note Terminal-Bench 75.1% in description, consolidate pill text
- #5 DeepSeek V3.2 → Qwen 3.6-Plus: leads Terminal-Bench at 61.6%, 88.2% GPQA,
1M context, available now on Alibaba Cloud + OpenRouter
Writing section (reordered based on EQ-Bench CW scores):
- #1 Claude Sonnet 4.6 (1936 EQ-Bench CW — highest, best voice consistency)
- #2 Claude Opus 4.6 (Mazur 8.53, IF Arena #1, 1M context for literary depth)
- #3 Gemini 3.1 Pro (Arena CW #1 1487, AI-tell avoidance, 2M context)
- #4 GPT-5.4 (noted as ~9th on Arena CW, better for structured/commercial writing)
- #5 Meta Muse Spark → Kimi K2.5 (/usr/bin/bash.60/.50, ~1700 EQ-Bench CW, live API)
Muse Spark removed — no commercial API available yet
Reasoning section:
- Gemini 3.1 Pro GPQA: 95.45% → 94.1% (more conservative/recent figure, consistent
with both agents' data)
- Added ARC-AGI-2 77.1% for Gemini 3.1 Pro (#1 on visual reasoning too)
- Opus 4.6: added note that Sonnet leads GDPval-AA (1633 Elo #1) for throughput
- #5 DeepSeek V3.2 → Qwen 3.6-Plus (88.2% GPQA, 1M context, same model as coding)
Quick picker:
- Creative writing: Opus → Sonnet 4.6 (EQ-Bench #1, 85% cheaper)
- Hard reasoning: 95.45% → 94.1%, add ARC-AGI-2 mention
- Budget pick: DeepSeek V3.2 → Gemini 3 Flash Thinking (/usr/bin/bash.50/1M, 89.8% GPQA)
Setup boxes:
- Self-hosted: Muse Spark → Qwen 3.6-Plus + Gemma 4 26B MoE (Apache 2.0,
82.3% GPQA with 3.8B active params, best edge/self-hosted reasoning)
Overall section: unchanged (top 5 still correct per both agents)
Search section: unchanged (no new data from either agent)
Every page now has a consistent hamburger menu on mobile (≤640px) that
lets users navigate the full site without ever getting stranded.
eli5/index.html:
- Added hamburger button (#eli5-hamburger) + mobile menu (#eli5-mobile-menu)
- Menu links: Home, Why/Features/Compare/Install, ELI5, Models, Community, Get started
- Hamburger CSS + open/close JS added
models/index.html:
- Added hamburger button (#models-hamburger) + mobile menu (#models-mobile-menu)
- Menu includes section divider between main site links and sub-pages
- Hamburger CSS + open/close JS added (placed before tab-bar scroll JS)
compare/*.html (all 9 pages):
- Added hamburger button (#compare-hamburger) + mobile menu (#compare-mobile-menu)
- Menu: Home, ← All comparisons, divider, ELI5, Models, Community, Get started
- Same hamburger CSS (scoped to each page's inline style block)
- Same open/close JS before </script>
community/index.html:
- Existing hamburger kept, mobile menu updated to include ELI5 and Models links
(was missing them — now links to ../eli5/ and ../models/ in accent color)
Behaviour: hamburger hidden at desktop (display:none), shown at ≤640px.
Clicking toggles .open class. Clicking outside closes. All menus are
position:fixed so they overlay content cleanly. Each menu uses a unique
id to avoid collisions when pages are eventually combined.
eli5/index.html:
- Logo: ☿ → ⚡ to match compare/community pages
- Nav: glass effect (rgba 13,17,23, backdrop-filter blur) matching site nav
- Light mode nav override (rgba 246,248,250)
- CTA: github URL → hermes-agent.nousresearch.com install docs
models/index.html:
- Logo: ☿ → ⚡ to match compare/community pages
- Nav: glass effect matching site nav
- Light mode nav override
- CTA: github URL → install docs
- Tab bar: position:sticky so it stays visible while scrolling
- BUGFIX: malformed CSS comment /* HOW TO PICK --> was breaking CSS
parser and causing picker-card, picker-grid, MOBILE rules to be
silently ignored — picker cards had no background, mobile breakpoints
not applying. Fixed comment to /* HOW TO PICK */
- eli5/index.html: visual ELI5 explainer with SVG diagram, before/after,
step-by-step breakdown, and concept cards for new users
- models/index.html: 2025 model guide with 5 categories (Overall, Coding,
Writing, Search, Reasoning), 25 model cards with benchmark data, tab
nav, quick picker, and setup instructions for all providers
- index.html: add ELI5 and Models links to desktop nav and mobile menu
- .gitignore: exclude __pycache__ and .pyc files
index.html:
- Added 'Community' link to desktop nav (last item after Resources)
- Added '🌎 Community' to hamburger mobile menu (highlighted in accent color)
- Added community-banner callout at the top of the Resources section —
a prominent amber-bordered card linking to community/ so it's the
first thing you see in Resources, not the last card
community/index.html:
- Added full hamburger menu (CSS, HTML, JS) matching the pattern from
the compare pages — appears on mobile at ≤640px, links back to all
main Hermes sections and the community page itself
- hermes-webui badge: 'Official' → 'Popular' (it's a community project,
not an official Nous Research product)
- hermes-webui description updated to remove 'The official Web UI' phrasing
Removed swarmclaw (github.com/swarmclawai/swarmclaw) and ClawMem
(github.com/yoloshii/ClawMem) — both named after or primarily built
around OpenClaw rather than Hermes. The community page should showcase
Hermes-native projects.
Two other projects (agency-agents-zh, avoid-ai-writing) merely mention
OpenClaw compatibility in passing alongside Hermes/Claude Code — those
are kept since they're genuinely Hermes-compatible tools.
Updated stats: 27 → 25 community projects.
* feat: add community ecosystem page with 30+ third-party Hermes projects
- New community/index.html page showcasing the Hermes ecosystem
- 6 categories: Curated Lists & Guides, Agent Roles & Configs,
Web Interfaces & Desktop, Skills & Plugins, Tools & Orchestration,
Memory Providers, Official Projects
- 30+ projects with 50+ star minimum, sorted by stars
- Featured badges for top projects (agency-agents-zh 5.8k, gbrain 4.8k,
orange-book 1.9k, awesome-hermes-agent 1.1k, hermes-workspace 1.1k)
- Matches existing site design: dark/light theme, card grid, responsive
- Added Community & Ecosystem link to main index.html resources grid
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: community page review corrections
- Theme toggle: ☾ → 🌙 in button HTML; \u263E → actual 🌙 emoji
in setTheme() JS. The ☾ text glyph (U+263E) renders as a thin line
and was the same regression seen in PR #270. Fixed to proper emoji.
- Stats bar: '30+ community projects' → '27 community projects' (actual count)
- Stats bar: '10+ categories' → '7 categories' (actual count)
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: QA polish — stale model refs, duplicate CSS, misplaced card, HTML fix
Model reference updates (across 5 comparison pages):
- GPT-5.4 → GPT-5.5 everywhere (chatgpt, claude-ai, claude-code, copilot, perplexity)
- o3-pro → o5-mini (chatgpt.html)
- Mistral → DeepSeek/GLM as provider examples (chatgpt, claude-ai)
- Claude Sonnet 4.6 → Claude Opus 4.6 (claude-code, copilot)
- "GPT-5 variants" → "GPT-5.5" (perplexity-computer)
index.html fixes:
- Remove duplicate .webui-layout > * { min-width: 0 } (3x → 1x)
- Remove duplicate .install-layout > * { min-width: 0 } (3x → 1x)
- Remove duplicate .compare-more CSS block (compact + expanded → single)
- Fix stray </div> in "why" section causing malformed HTML
- Fix theme toggle icon: sun shown in dark mode (switch-to), moon in light
- Move Perplexity Computer card from "Who it's for" to head-to-head cards
- Add Perplexity Computer as 5th head-to-head comparison card
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: correct model references -- GPT-5.4 is current, not 5.5
- Revert GPT-5.5 back to GPT-5.4 across all comparison pages
- Remove o5-mini (does not exist) -- replaced with generic wording or GPT-5.4 Pro
- Remove o3-pro reference (chatgpt.html)
- All model refs now accurate: GPT-5.4, Claude Opus 4.6, Claude Sonnet 4.6, Gemini, DeepSeek, GLM
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: correct review findings before merge
- index.html: revert \u263E back to 🌙 emoji in setTheme()
(U+263E is a text-mode glyph that renders as a thin line; this
was previously fixed in b4fa657 and accidentally reintroduced)
- chatgpt.html: 'GPT-5.4 Pro' → 'GPT-5.4' (invented product name)
- claude-ai.html: 'GLM' → 'Gemini' (GLM too obscure for audience)
- copilot.html: 'Claude Opus 4.6' → 'Claude Sonnet 4.6'
(GitHub Copilot offers Sonnet, not Opus, in its model picker)
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
index.html:
- CRITICAL layout bug: compare-more pills div was inside #why section
(an orphaned block from a previous failed insertion attempt), causing
the #why section to be 5705px tall and the pill bar to appear right
after the Why section header instead of after the Compare section.
Fixed by removing the stray block and placing compare-more correctly
inside #compare section, after the head-to-head cards.
- Added Perplexity Computer row to the main comparison table (was missing;
it had a card and pills but no row in the overview table).
Model reference updates:
- chatgpt.html: GPT-4o → GPT-5.4, o3 → o3-pro (two instances each)
- claude-ai.html: GPT-4 → GPT-5.4 in provider-flexibility bullet
- claude-code.html: 'GPT-4o for one thing' → 'GPT-5.4 for one thing,
Claude Sonnet 4.6 for another' in provider-agnostic routing example
- copilot.html: 'Claude, GPT-4o, Gemini' → 'Claude Sonnet 4.6, GPT-5.4,
Gemini' in model roster description
- cursor.html: 'valued at $29.3B' → 'valued at over $29B' (durable phrasing)
index.html:
- CRITICAL JS BUG: setTheme() had a ghost orphan 'else' block left from
the old hljs-switching version that was never fully removed. The function
correctly closed at line 2077 but a dangling '} else { ... hljsTheme.href
... }' remained at lines 2078-2082. This caused a JS syntax error that
silently killed the entire <script> block — theme toggle, hamburger menu,
copy buttons, nav active state, smooth scroll: all broken. Fixed by
removing the dead code entirely.
- Add overflow-x: hidden to html element to prevent any overflowing content
from enabling horizontal scroll on mobile.
compare/*.html (all 9 pages):
- Add overflow-x: hidden to html element (same fix, same reason)
- Add @media (max-width: 640px) block that hides .nav-back span text and
the 'Get started' CTA on narrow screens. The compare nav has three items
(logo, back-link, theme+CTA) with no hamburger — at mobile widths these
were colliding/overflowing. Now at ≤640px: back-link shows '←' only
(span hidden), CTA hidden, theme toggle stays visible and usable.
- Wrap nav-back text in <span> so the mobile rule can target it
- Add word-break: break-word and min-width: 0 to .narrative to prevent
long words/URLs from forcing horizontal overflow
New page: compare/perplexity-computer.html
Content sourced from official Perplexity docs, changelog, help center,
TechCrunch, and independent reviews. Covers:
- What Perplexity Computer actually is (cloud agentic workflow engine,
launched Feb 25 2026, GA to Pro in March 2026)
- Multi-model routing across 19+ frontier models (Claude Opus, GPT-5,
Gemini, Grok, etc.)
- Tasks scheduling feature (daily/weekly minimum; 10-40 task cap by plan)
- 400+ OAuth connectors + MCP support (added March 2026)
- Cloud-only constraint, data sovereignty trade-off, credit cost model
- Perplexity Personal Computer (Mac mini hybrid, waitlist only)
- Full feature-compare grid, 21-row comparison table, who-should-choose cards
Also:
- Added 'vs. Perplexity Computer' comparison card to index.html compare section
- Added compare pill to index.html and all 8 existing compare pages
The images used src="../images/" relative paths. On the local test server
(serving from the repo root) this resolved correctly. On GitHub Pages the
site is served at /hermes-webui/, so ../images/ resolves one level up to
/images/ (the GH Pages root), which doesn't exist — all 5 screenshots
were 404ing in production.
Fixed to src="images/" (same-directory relative), which resolves correctly
to /hermes-webui/images/ on GitHub Pages.
Caught during post-merge production browser test.
* docs: redesign GitHub Pages site with screenshots, updated content, and visual polish
- Add Web UI screenshot gallery with 5 images from README (hero + 4 thumbnails)
- Add clickable lightbox for full-size screenshot viewing
- Add "Who is Hermes for" section with 5 persona cards from HERMES.md
- Update all comparison content from HERMES.md PR #261 (accurate mid-2026 data)
- Update competitor details: Claude Code 26-event hooks, Cursor v3.0/Automations,
ChatGPT Agent Mode, Claude Cowork scheduling, OpenClaw 24+ platforms
- Add subtle grid pattern background to hero section
- Add gradient text effect on hero accent
- Add gradient top-border animation on pillar card hover
- Add convergence left-border accent callout
- Dark theme stays default; light theme polished
- Fix theme toggle: moon icon in dark mode, sun in light mode
- Add proper meta tags: og:image, twitter:card, theme-color, favicon, canonical URL
- Add Google/Gemini to LLM providers list
- Add voice input and mobile responsive to Web UI feature checklist
- Nav now includes "Who it's for" section link
- All responsive breakpoints updated for new sections
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: resolve all image, light-mode, and regression issues from PR #262
Image fixes:
- All 5 screenshot images were broken or using private GitHub auth-gated
user-attachments URLs. Downloaded and committed all images locally:
images/ui-hero.png, images/ui-light-mode.png, images/ui-settings.png,
images/ui-workspace.png, images/ui-sessions.png
- Fixed 4 gallery thumbnails to use unique images (was duplicating light-mode)
- og:image and twitter:image updated to absolute URLs for social card crawlers
Light mode code block fix (re-applied from previous gh-pages work):
- PR was branched from pre-fix state, so light-mode code block bug returned
- Code blocks always use dark bg (#0d1117) in both themes
- [data-theme=light] .code-wrap override now correctly stays at #0d1117
- setTheme() no longer switches hljs to github.min.css (caused dark-on-dark)
- code-header, copy-btn, code-lang use fixed dark-bg colors, not var(--text-muted)
which resolves to near-white in dark mode and near-black on dark backgrounds
Theme toggle emoji:
- Was using ☾ / ☾ (Unicode text crescent, renders as thin line)
- Now using 🌙 (proper moon emoji) for dark mode display
Grid overflow fix (re-applied):
- .webui-layout > * { min-width: 0 } re-applied
- .install-layout > * { min-width: 0 } re-applied
- Without this, <pre> code blocks in grid columns force columns beyond container
Comparison section (re-applied from previous gh-pages work):
- 4 comparison cards made clickable with onclick, role=link, card-arrow
- 8 compare-pill links to deep-dive pages added below head-to-head section
- Keyboard navigation (Enter/Space) for clickable cards
- Removed duplicate compare-more block and keyboard handler (applied twice due
to PR being based on pre-fix gh-pages)
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Pulls all screenshots from the README directly into the gh-pages branch
so they can be served as local assets (no auth-gated CDN dependencies).
Images sourced from:
- user-attachments (ui-hero, ui-light-mode, ui-settings) — downloaded
with browser headers since these require referrer to serve publicly
- master:docs/images/ (ui-workspace, ui-sessions) — extracted from git
Files:
images/ui-hero.png 2834x1734 783K Three-panel dark mode layout
images/ui-light-mode.png 2830x1742 786K Light mode with profile support
images/ui-settings.png 2822x1722 779K Settings panel / password config
images/ui-workspace.png 2822x1738 720K Workspace file browser
images/ui-sessions.png 2828x1732 633K Session management with tags
These are referenced by the PR #262 branch (docs/site-redesign-2026)
using relative paths (images/*.png). All 5 are Retina-res PNG
screenshots of the actual web UI.
Light mode fixes:
- Code blocks always use dark background (#0d1117) in both themes.
Previous behavior: setTheme('light') switched hljs to github.min.css
(light theme = near-black text) but kept code-wrap background dark,
producing invisible dark-on-dark text. Fix: always use github-dark
hljs + keep bg dark. Result: consistent readable code in both themes.
- code-header uses fixed dark colors (no longer rgba(255,255,255,0.04)
which was near-invisible on the dark bg it was compositing against).
- copy-btn uses fixed dark-theme colors for the same reason.
Index page updates:
- Comparison cards are now clickable links to deep-dive pages.
Card has onclick, role=link, tabindex=0, keyboard Enter handler,
arrow indicator (→) that animates on hover, and focus ring.
- Added compare-more pills row below the head-to-head cards linking
to all 8 comparison pages.
- CSS for .compare-pill, .compare-more, .card-arrow added.
New pages (compare/):
- compare/openclaw.html
- compare/claude-code.html
- compare/codex.html
- compare/opencode.html
- compare/cursor.html
- compare/copilot.html
- compare/claude-ai.html
- compare/chatgpt.html
Each page has: full CSS (light+dark themes), verdict banner, feature
comparison 2-col grid, 5-section narrative with official docs links,
full comparison table, who-should-choose section, nav pills to all 8
sibling pages, and CTA strip. Content drawn from HERMES.md (with the
/loop and scheduling corrections from PR #261 applied).
Layout bugs:
- Add min-width: 0 to .webui-layout children and .install-layout children.
CSS grid 1fr 1fr columns were expanding beyond the container because <pre>
code blocks inside don't wrap. This caused .webui-steps to be 842px wide,
overflowing the 1100px container by 64px (child right edge at 1254px,
container right at 1190px). min-width: 0 lets grid children shrink to
their fr allocation and overflow-x: auto on the code blocks handles the rest.
Content accuracy (matching HERMES.md fixes):
- Table: Claude Code 'Self-hosted scheduling' column was 'Cloud only' — now
correctly shows 'Cloud or desktop app' (desktop app scheduled tasks are real,
run locally on your machine with full file access)
- Comparison card: 'scheduling runs on Anthropic's cloud (your data leaves
your hardware)' was only half-true. Desktop app tasks stay local. Nuanced
to note cloud tasks are off-hardware; desktop app tasks run locally but
require the app to stay open.
nesquena-hermes