Skip to content
DocumentationTry Hanzo
Playground

Where to start

Six ways into Hanzo, and what each one is actually for. Some are a page you read, some are a command you run. None of them need a sales call first.

Pick what to try

Reach for the one that matches the problem in front of you.

Zen models

The family we train ourselves — 0.6B you can run on a laptop up to frontier. Most are open weight; read the specs and the price per million tokens.

See the models

Bot

Bots as backend services: routing, async runs that last hours, durable state, and an identity per bot instead of a shared key.

Read about Bot

MCP

Thirteen tools — shell, files, code, git, HTTP — behind one server. One block in .mcp.json and your client has them.

Set up MCP

Hanzo Dev

A coding agent in your terminal. It opens the files, makes the change, runs the tests, and shows you the diff.

Install Dev

Chat

One thread with every model, your MCP tools, your files and a code sandbox. Free to try at hanzo.chat.

Open Chat

Agents

A Python SDK where an agent is a model, instructions and tools — and several of them can sit behind one router.

Read the SDK

Or copy-paste a starter

A key, a model name, and a message is the whole first request.

curlOne request
curl https://api.hanzo.ai/v1/chat/completions \
  -H "Authorization: Bearer $HANZO_API_KEY" \
  -d '{"model":"zen5","messages":[{"role":"user","content":"Hello"}]}'
pythonAn agent
pip install hanzo-agent

from agents import Agent, Runner
agent = Agent(name="Assistant", instructions="You are a helpful assistant")
print(Runner.run_sync(agent, "Write a haiku about recursion").final_output)
bashTools for your editor
npm install -g @hanzo/mcp
hanzo-mcp list-tools
hanzo-mcp install-desktop

Read the code first if you want

Dev, MCP, the agent SDK, the CLI, the editor extensions and the clients in four languages are all on GitHub. So are most of the model weights.

Open source

License: Apache-2.0hanzoai

Get Playground

Hands-on sandbox for every product