Skip to content

zbin0929/CoreyOS

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

826 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

CoreyOS

CI English | 中文

A developer-first AI control plane built on Hermes Agent.

Chat across any model, manage industry Skill Packs, orchestrate workflows, and route across messengers — all from one keyboard-first macOS / Windows app.

Tauri 2 desktop app · 15 MB binary · 50–100 MB RAM · macOS 12+ / Windows 10+


Features

Chat & Multi-Agent

  • 🤖 Multi-model chat — streaming SSE, tool progress, markdown rendering, syntax highlighting. Switch between DeepSeek, GPT-4o, Claude, Qwen, Kimi, Gemini, and local Ollama from the composer.
  • 🔀 Multi-agent console — run multiple Hermes instances side-by-side, each with its own LLM profile, adapter, and budget.
  • 📎 File attachments — drag-drop, paste, or file picker. PDF/Word/Excel content extraction. Per-agent sandbox scopes.
  • 🔄 Compare models side-by-side — one prompt, N models in parallel.

Knowledge & Memory

  • 📚 Knowledge base (RAG) — upload documents, auto-chunk, search with Jaccard keyword matching or optional BGE-M3 semantic vector search with RRF fusion. Injected into every chat turn.
  • 🧠 Memory editorMEMORY.md + USER.md with syntax highlighting, capacity meter, FTS5 session search. Holographic fact store stats (fact count, categories, decay).
  • 🎙️ Voice — push-to-talk ASR + TTS (OpenAI, Zhipu, Groq, Edge TTS).

Workflow & Automation

  • Workflow engine — DAG-based workflow editor with conditional branching, loops, browser automation steps, and parallel execution.
  • Scheduler — cron prompts with output captured as Markdown.
  • Tasks & Approvals — track agent-initiated tasks, approve or deny dangerous operations.
  • 🛡️ File Guard — cross-platform file-ops-guard blocks destructive operations (delete/overwrite) on Desktop/Documents/Downloads with native confirmation dialog.

Skills & MCP

  • 🧩 Skills Hub — browse + install from 7+ community sources by wrapping the hermes skills CLI.
  • 📦 Skill Packs — industry-specific bundles (views, skills, workflows, MCP servers, schedules) loaded via manifest.yaml. White-label via customer.yaml.
  • 🔌 MCP server manager — stdio + URL transports, per-Pack MCP auto-registration, desktop-native tools (notifications, file picker, deep links).

Observability

  • 📊 Analytics — token usage, cost estimation, latency tracking, error rates, health radar, CSV export.
  • 🕸️ Trajectory — visualize past sessions as a timeline of messages and tool calls. Nested subagent delegation trees. Per-message token/latency stats.
  • 💰 Budgets — per-model / per-profile spend caps with 80% warnings.
  • 📜 Logs — real-time gateway + agent log viewer.

Platform & Channels

  • 📡 16 messaging gateways — Telegram, Discord, Slack, WeCom, WeChat, Feishu, Matrix, WhatsApp, Signal, DingTalk, Email, SMS, iMessage, Mattermost, Webhooks, Home Assistant.
  • 🧰 LLM profile library — define {provider, base_url, model, api_key_env} once on /models, reuse across agents. 11 pre-built provider templates including 6 domestic Chinese LLMs.
  • ⌨️ Keyboard-first⌘K palette, every page reachable by shortcut.
  • 🌏 中文 + English — full zh-CN localization; auto-detects locale.

Screenshots

Screenshots coming with v0.3.0 release.


Install

Download the latest release from GitHub Releases. For building from source, see SETUP.md.

Prerequisites (runtime)

  • Hermes Agent installed and reachable on its local gateway (usually http://127.0.0.1:8642). Install from hermes-agent.nousresearch.com. CoreyOS without Hermes runs in read-only stub mode — useful for tinkering but most features won't work.
  • Operating system: macOS 12+ / Windows 10+.

First run

Open the app → Home page. The onboarding checklist auto-detects:

  1. Connect to Hermes — green chip = gateway reachable. Settings page if not.
  2. Pick a model — any OpenAI-compatible provider, configured in Hermes itself.
  3. Start your first chat — send a prompt. That's it.
  4. Set up your profile — write one or two lines in Memory → User profile so the agent knows who you are.
  5. Connect a messenger (optional) — Telegram / Discord / etc. from Channels.

Press ⌘K anywhere to fuzzy-search across every page + runbook. Click the ? icon next to any page title for in-context help.


Pages

Page Description
Chat Multi-model streaming chat with tool progress, slash commands, attachments, and token tracking
Home Dashboard with onboarding checklist, Hermes status, and quick actions
Workflows DAG workflow editor with browser automation, conditionals, and scheduling
Tasks Agent-initiated task tracking with approval flow
Models LLM profile library — define once, reuse across agents
Skills Browse, install, and manage Hermes skills from 7+ sources
Knowledge Upload documents → auto-chunk → keyword/semantic search → inject into chat
Memory Edit MEMORY.md / USER.md, view holographic fact store stats
MCP MCP server manager — stdio + URL transports, desktop-native tools
Channels 16 messaging gateways (Telegram, Discord, WeCom, etc.)
Trajectory Session timeline — messages, tool calls, subagent trees, token/latency stats
Analytics Token usage, cost, latency, error rates, CSV export
Logs Real-time gateway + agent log viewer
Budgets Per-model / per-profile spend caps
Settings Provider config, profiles, voice, scheduler, runbooks, terminal, and more

Status

Milestone Status
Phase 0–12 (Foundation → File Intelligence) ✅ All shipped
v0.2.0 (White-label + Pack loader + 12 view templates + license + analytics) ✅ Shipped
v0.2.13 (Windows file-ops-guard fix, cross-platform dialog) ✅ Released
v0.3.0 (Cross-border e-commerce Pack) 🔧 In progress

Test totals: Rust 555 · Vitest 112 · Playwright 77 · all green.

Full roadmap in docs/05-roadmap.md; changelog in CHANGELOG.md.


Architecture

Tauri 2 desktop app (single binary, no Electron).

  • Rust (src-tauri/): 193+ IPC commands, SQLite session store, sandbox gate, Pack loader, workflow engine, MCP stdio bridge, knowledge base (Jaccard + BGE-M3 RAG), file-ops-guard, adapter layer for Hermes /v1/chat/completions.
  • TypeScript (src/): React 18 + TanStack Router, zustand stores, Tailwind + shadcn-style components, 12 Pack view templates.
  • Data: all state in ~/.hermes/ (config, skills, workflows, memory, MCP servers). No remote server.

More in docs/01-architecture.md.


Contributing

Start here:

Run pnpm tauri:dev:clean and start poking.


Name

CoreyOS is the project. The Rust crate is caduceus (legacy name, not worth renaming). The app bundle is Corey.


License

See docs/licensing.md for details.

About

Control plane for AI agents — Tauri + React

Resources

Stars

Watchers

Forks

Packages

 
 
 

Contributors