tinycrab
Developer Experience First

Spawn AI Agents in Seconds

The fastest way to get an AI agent running. One line of code, instant agent.

Get StartedI'm an AI Agent

Watch them spawn

$ tinycrab spawn worker-0

Each crab is an independent agent with its own workspace, memory, and tools.

Built for Speed

You're building something that involves AI agents. You need one running now.

  • •Testing how your app handles AI responses
  • •Prototyping an AI-powered workflow
  • •Building an AI-agent-first platform
  • •Creating multi-agent simulations
  • •Spawning agent swarmssoon

You don't need the most powerful agent. You need an AI agent running right now.

Three Ways to Spawn

Choose your interface. Same agent underneath.

SDK

For TypeScript/JavaScript projects

import { Tinycrab } from 'tinycrab';

const tc = new Tinycrab({ apiKey: process.env.OPENAI_API_KEY });
const agent = await tc.agent('worker');

const result = await agent.chat('Write a hello world script');
console.log(result.response);
// → I've created hello.py with a simple hello world script.
//   The file prints "Hello, World!" when run.

await agent.destroy({ cleanup: true });

CLI

For quick testing in terminal

$ npm install -g tinycrab

$ tinycrab spawn my-agent
# → Agent 'my-agent' spawned
# → Server running on port 9000

$ tinycrab chat my-agent "Create a README"
# → [my-agent]: I've created README.md with project
#   documentation including installation and usage.
# → (session: abc123)

$ tinycrab cleanup my-agent
# → Agent 'my-agent' cleaned up

Docker

For CI/CD or cloud deployment

$ docker run -p 8080:8080 \
  -e OPENAI_API_KEY=sk-xxx \
  ghcr.io/jt-wang/tinycrab
# → Agent server running on port 8080

$ curl -X POST localhost:8080/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "Hello"}'
# → {"response": "Hello! How can I help you?",
#    "session_id": "session-a1b2c3..."}

What You Get

Everything you need, nothing you don't.

Full coding capabilities

bash, read, write, edit, grep, find

Isolated workspaces

Each agent has its own directory

Session persistence

Multi-turn conversations with context

Cross-session memory

remember/recall for persistent knowledge

8 LLM providers

OpenAI, Anthropic, Google, and more

Self-spawning

Agents can spawn new agents via bash

Works Everywhere

Same code, same behavior. Test locally, deploy anywhere.

docker run -p 8080:8080 -e OPENAI_API_KEY=sk-xxx ghcr.io/jt-wang/tinycrab
Deploy Guide

Ready to spawn?

npm install -g tinycrab && tinycrab spawn my-agent

Coming soon: Agent Swarm - coordinate multiple agents working together