AI Agent Framework
BearClaw
AI agent framework with defense-in-depth security, multi-agent orchestration, and provider abstraction. Zero SDK dependencies — only 2 runtime deps total.
npm install -g bearclawWhy BearClaw
Most AI agent frameworks bolt security on as an afterthought. BearClaw treats it as a core architectural concern — with encrypted secrets, rate limiting, SSRF guards, approval workflows, and policy engines built in from the start.
- Security-first with defense-in-depth design
- Multi-provider: Anthropic, OpenAI, Ollama, CLI delegation
- Multi-agent orchestration with team-based routing
- Zero SDK dependencies — all providers use fetch() directly
Built for security & control
API keys encrypted at rest with ChaCha20-Poly1305, sliding window rate limiting scoped per-agent and per-tool, SSRF protection, and configurable autonomy levels from locked to full.
Quick start
Install globally and start chatting with your configured AI agent.
$ npm install -g bearclaw
$ bearclaw
BearClaw looks for its config at ~/.bearclaw/config.json. On first run it encrypts your API keys in-place and generates a secret key.
What you get out of the box
Multi-provider LLM
Anthropic, OpenAI, Ollama, and CLI delegation with streaming support and exponential backoff retry.
Multi-agent orchestration
Team-based routing, @agent mention parsing, conversation tracking with fan-out/fan-in patterns.
Defense-in-depth security
SecurityPolicy, PolicyEngine, SSRF guard, rate limiting, approval workflows, and encrypted secrets.
Tool system
9 built-in tools with JSON Schema validation, before/after hooks, parallel execution, and structured results.
Multi-channel messaging
CLI REPL and Telegram with a unified message bus. HTTP gateway with pairing-based authentication.
Zero SDK dependencies
All provider integrations use fetch() directly. Only 2 runtime dependencies total.
Learn more
Check out the full README, configuration docs, and source code on GitHub.