Captain Claw
Command a fleet of AI agents — on every deck, every desk, every wrist
Most AI tools give you one agent in a chat box. Captain Claw Flight Deck is a multi-agent command center: spawn specialist teams, run seven orchestration modes, ship software with Code (plan → build → independent review), and compose deterministic Flows. Self-hosted, MIT licensed, and it works with every major LLM provider, or 100% local with Ollama. Since 0.8.2 your Flight Decks work together: a planner on your VPS asks a researcher on the Raspberry Pi at home, and you watch the request cross a live map. Your team works from a kiosk; you can just talk to a watch.
🌐 Website · 🚀 Live Demo · 📖 Docs · 📦 PyPI · 🧭 Guides · 📝 What's new
New in 0.8.3 — Plugged In, On the Record: connect any system to an agent, let agents read your own WhatsApp, and see every tool call they make. Release notes
Why Captain Claw
- 🛸 A fleet, not a chatbot. Spawn specialists from 31 ready-made archetypes, or describe a goal and let Agent Forge design the whole team: roles, models, tools and operating procedures.
- 🧩 The right shape for every problem. Seven orchestration modes, from blind ensembles merged by learned reliability to a collaborative blackboard, moderated councils, and a finisher that stops only when an independent judge says the work is done.
- 🛰️ Decks that work together. Flight Decks on a VPS, a laptop and a Raspberry Pi behind home NAT lend each other their agents over BotPort, and you watch every request cross a live map.
- 💻 An engineering department. Plan → your approval → build → three independent reviewers → fix, every phase a git commit in your own repo.
- 👥 Ready for a team. A kiosk your non-admins live in, an admin console, invitations by email, per-user Google and Microsoft accounts, shared agents and daily plan limits.
- 🔌 Plugs into anything. Give your CRM, shop or monitoring its own key and it talks to an agent over a WebSocket, HTTP or signed webhooks, or Flight Deck dials out to it. MCP works both ways: Claude drives your fleet, and your agents use any MCP server.
- 📡 Wherever you are. Web, WhatsApp (the Cloud API, or your own number), Telegram, a HUD for Meta Ray-Ban Display, three push-to-talk M5Stack devices, and Claude itself over MCP, all one conversation.
- 🧬 Memory that keeps thinking. Six memory layers shared across sessions and agents, and dream cycles that surface connections you didn't ask for.
- 🛡️ Agents you can check on. Every tool call lands in a tamper-evident register you and your admins can read, guards run before, during and after every action, and a fabrication guard catches a claimed send or calendar change no tool made.
- 🔓 Yours, end to end. Self-hosted, MIT, bring your own keys or run 100% local with Ollama. No lock-in.
Quick start
pip install captain-claw
flight-deck # the multi-agent dashboard: http://localhost:25080
First run starts onboarding. For a single agent in the terminal or the browser, Docker, a deck for your team or a voice device, see Getting started.
🔌 New in 0.8.3 — plugged in, on the record
- Agent links. Connections → Systems gives an outside system its own key to one of your agents: a WebSocket, plain HTTP or signed webhooks, files both ways. Or Flight Deck dials out to an orchestrator, so an agent behind a router needs no public address.
- Your own WhatsApp (experimental, opt-in). Link your number as a linked device; agents read your chats, files and voice notes, send only to chats you whitelist, and answer in the chats you pick when someone calls them by name.
- Every tool call, on the record. An append-only, HMAC-chained register of every call every agent makes: Activity for you, Audit for admins, and an
activity tool so an agent can tell you what it really did overnight.
- Nothing waits unseen. Nudges and proposals that need your approval reach WhatsApp, Telegram, your devices and the bell, with Approve and Reject right there.
- And: idle agents sleep and wake when needed, Shared folders with OneDrive in the kiosk, Word documents with real tracked changes, and six new tools.
→ Release notes · Agent links · WhatsApp Web · Activity and Audit
🛰️ BotPort — Flight Decks that work together since 0.8.2
Every Flight Deck used to be an island. Now any number of them join one network through BotPort, a small hub that is happy on a Raspberry Pi in a tmux pane. Agents address each other as Agent@deck: your planner on the VPS asks Researcher@pi at home, and the answer streams back with live progress, as if both sat on the same deck.
- Every deck dials out. A Pi behind home NAT or a laptop joins as easily as a server.
- Add a deck in a minute, with no restart. Paste the join code from the hub's Decks page into that deck.
- Everyone can share the agents they own, after a dialog that says plainly what other decks can and can't do with them.
- Safe by construction. A request from another deck runs in a lane of its own: no email, no Flows, no fan-out, and your deny list applies.
→ BotPort and MCP guide · Set it up, step by step
🧭 The kiosk — a front door your whole team walks through
The chat-first view your non-admins live in is a complete place to work: one-click "What can I do?" skill cards, a File library and a Data library that summarise everything the agents keep, Shared folders with Google Drive and OneDrive beside them, Scheduled tasks that deliver to chat, email or Telegram, one timeline per agent with named topics, and a ⋯ menu that shares an agent on the network. Admins get an Admin console, invitations by email, a waitlist, branding and daily plan limits.
→ The kiosk and simple layout · Running a deck for a team
🎙️ Devices — talk to your agents, they talk back
Push-to-talk clients on three M5Stack devices: the round-AMOLED StopWatch, the 4" colour e-paper PaperColor and the StickS3. Press, talk, and the device speaks the answer sentence by sentence while the model is still writing it. Voice notes wait offline until the deck is back, every question carries the room's context, and updates arrive signed, over the air, from your own deck.
→ Devices and channels, with WhatsApp, Telegram, the smart-glasses HUD and agents in your inbox.
💻 Code — an engineering department, not a coding assistant
Describe what you want built. A router sizes the job; big work runs plan → your approval → build → three independent reviewers in parallel → triage → capped fix loop, every phase a git commit in the folder's own repo. Agents query a persistent Code Map instead of re-reading your tree, per-turn token costs are shown in chat, and the whole run exports as one auditable Markdown transcript.
Cursor and Claude Code give you one very good agent that plans, writes, reviews, and tests in a single context. Code splits those roles across independent agents — the reviewer never grades its own homework, QA actually runs the tests — and puts you at the approval gate before anything is built. A complete real run (an RTS game with 185 tests, ~$2.50 on DeepSeek) ships in examples/code/.
→ Code in the Flight Deck guide
Pick the right shape for the problem
One agent isn't always the answer — and one orchestration strategy never is. Captain Claw ships seven distinct modes, each tuned to a different kind of work.
| |
|---|
| 🛠️ Agent Forge — describe a goal, an LLM designs the whole team (roles, models, tools, SOPs) | 🏛️ Agent Council — specialists deliberate across moderated rounds, then vote and synthesize |
| |
| 🔀 Basna — N agents answer the same question blind, merged by reliability into one high-confidence result | 🤝 Vatra — a team collaborates on a shared blackboard, each owning sections, improving over review rounds |
| |
The fifth mode is 💻 Code (above). The sixth is Flight Deck itself — spawn, monitor, and coordinate the whole crew from one dashboard. The seventh is 🦇 Bat, the stubborn finisher: checkpointed steps with retries, survives restarts, and stops only when an independent, fail-closed judge says the task is done.
→ All seven modes
More than orchestration
| |
|---|
⚙️ Flows — a deterministic automation DSL (triggers, gosub, parallel spawn/join, on error), validated by a parser, not interpreted by an LLM | 🔭 Observatory — it keeps thinking when you're away: a live, auditable stream of thoughts, dreams, and standing intentions |
| |
- Plan Mode & Deep Mode — for when "probably right" isn't good enough: a reviewable DAG of verified steps, and frontier-quality answers via multiple rollouts + self-consistency voting + diverse-lens critics.
- 6-layer shared memory — working, semantic (vector + BM25), deep (full-text), insights, autonomous nervous system, and self-reflection, carried across sessions and across every agent. → Memory and cognition
- Shared Virtual File System — a host-sandboxed, per-user workspace the whole fleet reads and writes, with per-file authorship.
- Every model — OpenAI (API key or Sign in with ChatGPT), Anthropic, Google Gemini, DeepSeek, OpenRouter, Ollama and on-device LiteRT, mixed freely per agent. → Inside an agent
See it in action
📚 Documentation
| Guide | What's inside |
|---|
| Getting started | Install and the launchers, a deck for a team, a voice device on your desk, Docker, configuration |
| Flight Deck | The command center: seven orchestration modes, Code, Plan and Deep mode, Flows, the kiosk, activity and audit, running a deck for a team |
| BotPort, MCP and agent links | Flight Decks that work together, Flight Deck as an MCP server for Claude, MCP servers for your agents, connecting your own systems |
| Devices and channels | StopWatch, PaperColor and StickS3, the smart-glasses HUD, WhatsApp (and your own WhatsApp), Telegram, notifications, agents in your inbox |
| Memory and cognition | Six memory layers, the context engine, dreaming, the Brain Graph |
| Inside an agent | Every supported model, the 63 built-in tools, safety guards, the agent web UI |
| Architecture | Where each component lives in the code |
| What's new | Every release, newest first |
Reference: USAGE.md (every command, tool and config key) · FLOWS.md (the Flow language) · Team deployment · BotPort deployment · Glasses HUD · Device firmware · Release notes · All guides
⭐ Star the repo
Captain Claw is MIT licensed and free forever — your data stays on your machine, bring your own keys or run 100% local with Ollama. It's built solo and growing fast, and every star helps another builder discover it. If it looks useful, would you give it a star?
License
MIT
