Overview
Agenthood is a multi-agent framework — a whole crew of AI specialists. Each member has a defined role, trigger condition, and behavior. They enforce engineering standards, review code and commits, manage memory, and run autonomously via the TypeScript runtime or inside Claude Code.
Stack: TypeScript · Claude AI · Groq/OpenAI/Ollama/OpenCode · LanceDB · Tree-sitter · VS Code
Links: agenthood.flabs.tech · GitHub · npm
The Society
All members, sorted by craft:
| Category | Members |
|---|---|
| Engineering | Architect, Tester, Debugger, Strategist |
| Validation | Reviewer, Auditor, Doorman, Sentinel, Warden |
| Lifecycle | Scribe, Herald, Envoy, Steward, Operator |
| Knowledge | Librarian, Oracle |
Browser Studio
The Studio playground is a browser-based chat interface for the whole crew. Built with Next.js 16, Tailwind CSS 4, and the agenthood TypeScript runtime:
- Select any member and configure provider/model
- SSE-streamed responses with provider failover
- 6 providers: Anthropic, OpenAI, Groq, Ollama, OpenCode Zen/Go
- Server-side rate limiting, auth, and structured logging
- Conversation persistence in localStorage
TypeScript Runtime
Run agents autonomously via the CLI:
npx agenthood run the-scribe "write a commit message"
npx agenthood run the-reviewer "review the open PR"
npx agenthood run the-architect "plan implementation for issue #42"
The runtime features LLM provider routing with automatic failover, circuit breaker, 5-tier memory architecture, Agentic RAG with LanceDB vector store, and ReAct reasoning loops.