██████╗ ██████╗████████╗ ██████╗ ███████╗ ██╔═══██╗██╔════╝╚══██╔══╝██╔═══██╗██╔════╝ ██║ ██║██║ ██║ ██║ ██║███████╗ ██║ ██║██║ ██║ ██║ ██║╚════██║ ╚██████╔╝╚██████╗ ██║ ╚██████╔╝███████║ ╚═════╝ ╚═════╝ ╚═╝ ╚═════╝ ╚══════╝
English · 中文
Like an octopus — 9 brains, every arm thinks independently, but they share one brain.
Octos is an open-source AI agent platform built in Rust. It turns any LLM into a multi-channel, multi-user intelligent assistant — deployed as a single self-contained binary with zero runtime dependencies.
Connect your LLM API keys and messaging channels. Octos handles conversation routing, tool execution, memory, provider failover, and multi-tenant isolation. Manage hundreds of AI agent profiles through a web dashboard or REST API.
| Repo | Description |
|---|---|
| octos | Core platform — Rust binary, 16 LLM providers, 14 channels, DOT pipeline engine, multi-tenant gateway, octos serve web/API backend |
| octos-web | Web client — React SPA embedded at /app/ (chat, Slides Studio, Sites, Voice), talks to octos serve over the UI Protocol |
| octos-tui | Terminal client — Rust TUI over the UI Protocol (live streaming, approvals, diffs, onboarding) |
| octos-hub | Community skill registry — install and share agent skills |
- API-first — a REST + WebSocket UI Protocol backend (
octos serve). Build any frontend on top: web, terminal, mobile, CI/CD - Multi-tenant — 200+ profiles on 16GB. Each profile is an isolated OS process. Family Plan sub-accounts
- 16 LLM providers — Anthropic, OpenAI, Gemini, Vertex AI, OpenRouter, DeepSeek, Moonshot/Kimi, Qwen, MiniMax, Zhipu/GLM, Z.AI, Groq, NVIDIA, Ollama, vLLM, r9s — plus any OpenAI/Anthropic-compatible endpoint via
base_url - Multi-LLM pipelines — DOT graph workflows. Per-node model selection. Dynamic parallel fan-out
- Adaptive routing — 3-layer failover with Hedge racing, Lane scoring, and circuit breakers
- Broad tool set — shell, files, search, web, browser, MCP, office, spawn/cron — the full enabled tool set is sent to the model every turn
- 14 channels — Telegram, Discord, Slack, WhatsApp, Feishu/Lark, DingTalk, WeCom, WeChat, Matrix, LINE, QQ Bot, Email, Twilio (SMS), Web/API — plus the CLI
- Automatic memory — capture → extraction → consolidation into long-term memory, plus episodic (redb) and session (JSONL + LLM compaction) layers
- 5 queue modes — Followup, Collect, Steer, Interrupt, Speculative — per-session agent concurrency control
- Native office suite — PPTX/DOCX/XLSX via pure Rust (zip + quick-xml)
- Sandbox isolation — bwrap + sandbox-exec + Docker.
deny(unsafe_code)workspace-wide
# One-line installer — installs the binary + sets up `octos serve` as a service (macOS / Linux)
curl -fsSL https://github.com/octos-org/octos/releases/latest/download/install.sh | bash
# Homebrew (macOS Apple Silicon, Linux x86_64/ARM64)
brew install octos-org/tap/octos
# npm (macOS Apple Silicon, Linux x86_64/ARM64, Windows x64)
npm install -g @octos-org/octos# Windows (PowerShell)
irm https://github.com/octos-org/octos/releases/latest/download/install.ps1 | iexOr build from source with the full feature set (REST API + dashboard + channels):
cargo install --path crates/octos-cli \
--features "api,telegram,discord,dingtalk,whatsapp,feishu,twilio,wecom,wecom-bot,audio_mp3"export DEEPSEEK_API_KEY=your-key
octos chat --provider deepseek # interactive CLI
octos serve # web dashboard + REST API → http://localhost:50080Built with Rust. Powered by 9 brains.
