hermes-ui

Name: hermes-ui
Author: pyrate-llama

Verified

The command center for Hermes Agent — chat, steer, browse files, manage skills, and monitor everything from a single glassmorphic HTML app.

199stars

27forks

HTML

Installation

# Add to your Claude Code skills
git clone https://github.com/pyrate-llama/hermes-ui

Getting Started

Guides for using ai agents skills like hermes-ui.

Caveman: Cut Claude Token Use by 65%
How agent-side prompt compression works, when to use it, and when not to.
What is an AI Skills Marketplace?
Definitions, how marketplaces work, and how to choose between them in 2026.
Getting Started with AI Skills

Security ReportVerified

Last scanned: 5/30/2026

{
  "issues": [],
  "status": "PASSED",
  "scannedAt": "2026-05-30T16:11:07.113Z",
  "npmAuditRan": true,
  "pipAuditRan": false
}

Compare with similar skills

See how hermes-ui compares with popular alternatives.

hermes-ui vs hermes agent

See comparison

hermes-ui vs ecc

See comparison

hermes-ui vs n8n

See comparison

README.md

Frequently Asked Questions

What is hermes-ui?

hermes-ui is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by pyrate-llama. The command center for Hermes Agent — chat, steer, browse files, manage skills, and monitor everything from a single glassmorphic HTML app. It has 199 GitHub stars.

Is hermes-ui safe to use?

Yes. hermes-ui passed SkillsLLM's automated security scan — a dependency vulnerability audit plus prompt-injection heuristics — with no high-severity issues. You can read the full report in the Security Report section on this page.

How do I install hermes-ui?

Clone the repository with "git clone https://github.com/pyrate-llama/hermes-ui" and add it to your Claude Code skills directory (see the Installation section above).

What programming language is hermes-ui written in?

hermes-ui is primarily written in HTML. It is open-source under pyrate-llama on GitHub, so you can review or fork the full source.

Are there alternatives to hermes-ui?

Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh hermes-ui against similar tools.

Agentic AI for Beginners

Build your first AI agent from scratch - tool use, ReAct pattern, memory, deployment

41 minBeginner

Comments (0)

to leave a comment.

No comments yet. Be the first to share your thoughts!

Related Skills

ECC

by affaan-m

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

262,304

axi-front-design-skill lazyskills

Hermes UI

English · 简体中文

The command center for Hermes Agent — chat, steer, browse files, manage skills, and monitor everything from a single glassmorphic HTML app.

Built as a single-file HTML application with React 18, Hermes UI provides a full-featured chat interface, real-time log streaming, file browsing, memory inspection, and more — all through a lightweight Python proxy server.

Chat

Chat with inline image generation and live terminal panel Chat with split log panel and streaming response visibility

Tasks / Live Work

Tasks board showing active tool work and background follow-up tracking

Dashboard

Dashboard with live stats, recent activity, and installed skills

Skills Browser

Browse and manage installed Hermes skills

Cron Jobs

Scheduled cron jobs with edit, pause, and run-now controls

MCP Tools

Connected MCP servers grouped by category

File Browser

Live file browser for the active workspace with inline preview

Terminal

Built-in terminal area with Hermes and Claude Code tabs

What's new in v3.3.26

Reasoning Notes memory safety

Reasoning streams stop ballooning — cumulative reasoning snapshots are normalized into deltas before they reach the visible Reasoning Notes panel
Live notes stay bounded — long reasoning output is capped to the latest notes for display and storage, preventing unbounded browser RAM growth
Streaming saves are lighter — live reasoning notes no longer force full conversation mirror writes on every reasoning event; final transcript sync still preserves the completed turn
Regression coverage added — tests now cover cumulative reasoning snapshots so they do not append repeated full copies

What's new in v3.3.25

Large chat recovery and browser memory

Sidebar history stays compact — the browser stores chat metadata instead of full transcripts, avoiding localStorage and tab-memory blowups on large histories
Chats hydrate on demand — opening a chat loads its full transcript from the local server so refreshes do not pull every conversation into the browser at once
Refresh races are guarded — compact or partial browser saves can no longer overwrite full server-side chat mirrors during startup
Update button handles release checkouts — Update & Full Restart can now advance Hermes UI itself when the local install is checked out at an older release tag
Stress coverage added — browser tests now cover refreshing with large chat lists and switching into an inactive compact chat

What's new in v3.3.24

Workspace and updater polish

Spaces is no longer a dead end — non-chat screens now show a clear Chat return button in the header
Versions stay visible outside chat — Spaces and other tool screens show compact Hermes Agent and Hermes UI version chips
Agent updates handle release checkouts — Update & Full Restart can advance a detached Hermes Agent release tag and refresh the editable install metadata
Update failures are visible — failed repo updates now stop before restart and show the underlying pull/install output in Settings

What's new in v3.3.23

Agent workspace clarity

Top-bar workspace renamed — the folder chip now clearly reads as the active Agent Workspace instead of looking like the WebUI runtime directory
Useful path display — the chip shows a shortened workspace path alongside the saved Space selector
Manage shortcut — the chip includes a direct manage action into Spaces so users can add or change workspace folders intentionally

What's new in v3.3.22

Release reliability

Truthful update checks — /api/version compares published Releases with repository tags, so a delayed Release cannot leave the UI reporting an older version
One version source in the UI — desktop and composer footers now display the backend version instead of stale hardcoded text
Automated releases — pushing a v* tag publishes generated GitHub release notes
CI safety net — Python checks, release consistency, sensitive-data scanning, and a real Chromium startup smoke test run on pushes and pull requests

What's new in v3.3.18

Crash recovery context repair

Latest user turns stay visible — recovery context now preserves the newest user instructions even when tool/status chatter floods the transcript tail
Older workflows stay cancelled — the model prefix now tells Hermes that recent user corrections override stale transcript and memory when they conflict

What's new in v3.3.16

Sidebar ordering hotfix

No stale conversation cache — startup now bypasses browser cache when loading the sidebar conversation list
Always sorted state — conversation updates are normalized through the activity sorter so recently active chats stay at the top after refreshes and state rewrites
API cache headers — JSON responses now send no-store headers so browser resets do not reuse old sidebar data

What's new in v3.3.15

Cancellation + sidebar ordering

Cancel cleanup — cancelled work no longer leaves persistent composer status pills or repeated cancel messages in chat history
Context safety — cancellation markers are filtered out of model context so future turns do not treat them as assistant replies
Recent chat ordering — the sidebar now merges browser and server conversation caches before sorting, so the most recently active chats stay on top after refreshes

What's new in v3.3.14

Sidebar stability

Conversation dedupe — duplicate sidebar rows now collapse by session id on load, save, and live state updates
Refresh cleanup — persisted duplicate rows are repaired by the server instead of reappearing after reload

What's new in v3.3.13

Discussion feedback fixes

Firefox performance mode — disables decorative animations, background overlays, and backdrop filters in Firefox to avoid pinning a CPU core
Model picker cleanup — deduplicates model choices, corrects accidental gtp-* entries, and makes custom model selection apply reliably
Reasoning + slash menu polish — preserves spaces in streamed reasoning text, includes plugin skills in slash commands, and improves Dawn theme contrast

What's new in v3.3.11

Hermes UI 3.3.11 fixes stale compact context that could make an active chat look reset.

Session continuity

Stale context repair — if context_messages has fewer user turns than the saved transcript and no real compression marker exists, Hermes rebuilds from the full transcript
No false reset — chats with dozens of visible prompts no longer start the next turn from a two-prompt backend tail
Health counts fixed — session health now reports context from the repaired source instead of stale compact state

What's new in v3.3.10

Hermes UI 3.3.10 aligns normal chat context handling with the reference UI.

Context continuity

No invisible tail-trim — normal turns no longer narrow model history to a small local window before Hermes Agent sees it
Agent-owned compression — real context compression is left to Hermes Agent, which emits the visible compression event
Repair keeps substance — browser/server transcript repair rebuilds provider-safe history without dropping older visible turns

What's new in v3.3.9

Hermes UI 3.3.9 adds video uploads for local/video-capable model workflows such as Kimi.

Video uploads

Composer video attach — the file attachment button and drag/drop now accept common video formats
Workspace upload path — videos are saved into the active workspace's uploads/ folder and attached to the prompt by local path
Binary-safe prompts — video bytes are not inlined into chat history; Hermes receives the saved file path, MIME type, and size

What's new in v3.3.8

Hermes UI 3.3.8 fixes silent context narrowing in short chats with large tool receipts.

Reference-aligned context selection

Full transcript first — chats under 40 user prompts rebuild model context from the full backend transcript instead of stale context_messages
No silent compaction — bulky tool/output blobs are truncated for provider safety, but user/project turns stay in context unless real compaction happens
Detour recovery — after side quests like skill edits, Hermes keeps the original project prompts in model context

What's new in v3.3.7

Hermes UI 3.3.7 makes recovered UI tool receipts visible to model context without pretending they are provider-native tool calls.

Recovered tool receipt context

Provenance-labeled receipts — recovered UI toolCalls are shown as UI-observed tool events, not forged provider-native tool_calls
Bad correction pruning — model context drops broken self-corrections like "I didn't run any tools" when earlier UI-observed tool receipts exist
Verification guard — if UI receipts lack captured results, Hermes must verify the artifact/status with tools before claiming final state

What's new in v3.3.6

Hermes UI 3.3.6 fixes the browser/server repair path that could drop recent UI tool receipts after a transcript drift repair.

Tool receipt recovery

UI tool receipts restored — /api/session/health can now recover browser-persisted toolCalls from ~/.hermes/ui-conversations.json when the se