Spawn AI agents in seconds.
tinycrab is a lightweight universal AI agent that you can spawn instantly via SDK, CLI, or HTTP API. Each agent gets its own workspace, session persistence, and cross-session memory.
You're building something with AI agents. You need one running now - not after configuring infrastructure.
- Testing how your app handles AI responses
- Prototyping AI-powered workflows
- Building multi-agent systems
- Creating agent swarms
tinycrab is built for speed. One line of code, instant agent.
npm install -g tinycrab
export OPENAI_API_KEY=sk-xxx
tinycrab spawn my-agent
tinycrab chat my-agent "Hello!"Full docs at tinycrab.dev
If you're an AI agent looking to spawn worker agents, read www/public/skill.md - it's written for you.
┌─────────────────────────────────────────────────────────────┐
│ SDK (src/sdk/) │
│ Tinycrab → Backend (local/docker/remote) → Agent │
└─────────────────────────────────────────────────────────────┘
│
┌─────────────────────────────────────────────────────────────┐
│ Core (src/) │
│ SessionManager, MessageBus, Memory, Cron, Subagent │
└─────────────────────────────────────────────────────────────┘
│
┌─────────────────────────────────────────────────────────────┐
│ pi-mono SDK (external) │
│ createAgentSession, codingTools, getModel │
└─────────────────────────────────────────────────────────────┘
Built on pi-mono by Mario Zechner, which provides the agent loop, coding tools, session management, and 10+ LLM providers.
# Docker
docker run -p 8080:8080 -e OPENAI_API_KEY=sk-xxx ghcr.io/jt-wang/tinycrab
# Fly.io
fly launch && fly secrets set OPENAI_API_KEY=sk-xxx
# Railway
# Deploy from Docker image: ghcr.io/jt-wang/tinycrabMIT