Coding agents
Keep the coding agent you use. Run it with hanzo --<agent>: it hands work to Enso and Kai through the Hanzo toolset, and --cloud moves its model calls onto your Hanzo plan.
Nothing to port. The agent you use now keeps its login, its settings and its
sessions; hanzo starts it, attaches the Hanzo toolset for that run, and, with
--cloud, points its model calls at api.hanzo.ai.
curl -fsSL hanzo.sh | sh
hanzo login
hanzo --claude # or --codex, --cursor, --gemini, --qwen, --opencode, --goose, --hermes, …From each kind of setup
| You run today | Run instead | What changes |
|---|---|---|
| Claude Code on a Claude plan | hanzo --claude | The toolset is attached; Enso and Kai take work off your Claude plan |
| Claude Code through a router (claude-code-router, a LiteLLM proxy) | hanzo --claude --cloud | Model calls go to api.hanzo.ai; no proxy to run |
| Codex on a ChatGPT plan | hanzo --codex | The toolset is attached through -c mcp_servers.* |
| OpenCode, Kilo Code, Crush, Goose, Aider, Qwen Code, Hermes, OpenHands or Pi on OpenRouter or a provider key | hanzo --<agent> --cloud | Hanzo sets the provider for the run; your config is untouched |
| Cursor, Antigravity, Copilot, Amp, Gemini CLI, Grok Build, Cline, Kimi, Letta, Continue | hanzo --<agent> | Starts it on its own account; the toolset where the agent takes one |
Every agent, and what Hanzo adds to each →
Taking work off your plan
The toolset gives the agent llm (Enso), kai_decide (Kai), and Hanzo's code
and web search. Work handed to them is billed to your Hanzo plan, not the agent's,
and the summary at the end of each session lists every hand-off with the model
that took it and the tokens it used. What a session used →
Was this page useful?