Any agent.
Our rails.
Gen runs the coding agent you already use — Claude Code, Codex, Cursor and more — with Hanzo’s tools attached and every session linked to your Hanzo account: on Hanzo’s models, metered, or on the agent’s own account.
Hanzo Dev, our agent →Install
One line, then any agent.
Linux and macOS. hanzo.sh downloads gen’s native build and checks it: its checksum always, its provenance with a signed-in gh. Run it again to upgrade.
curl -fsSL https://hanzo.sh | sh -s gen # gen alone
curl -fsSL https://hanzo.sh | sh # or every Hanzo tool, gen includedgen # Hanzo Dev, the default
gen --claude # Claude Code; any agent is a flag
gen --codex "add a test" # a task, run headlessSign in once with the Hanzo CLI, hanzo auth login, and every session is linked to your account.
Gen and Dev
One runs agents. One is an agent.
The harness. Any agent, our rails.
It launches the coding agent you name with Hanzo’s tools attached, routes its model calls through Hanzo when you ask, and links the session to your account.Our agent.
The coding agent Hanzo builds, with Kai deciding each step. It is the agent Gen opens when none is named.Agents
The agent you use is a flag.
Each launches as itself, with its own interface and its own habits, and Gen around it. gen --help lists every one.
- Hanzo Dev--dev
- Claude Code--claude
- Codex--codex
- Cursor--cursor
- Antigravity--agy
- Gemini CLI--gemini
- OpenCode--opencode
- Kilo Code--kilo
- Crush--crush
- Pi--pi
- GitHub Copilot CLI--copilot
- Amp--amp
- Grok Build--grok
- Goose--goose
- Aider--aider
- Hermes Agent--hermes
- Cline--cline
- Continue CLI--continue
- Oh My Pi--omp
- Letta Code--letta
- Kode--kode
- OpenHands--openhands
- Open Interpreter--interpreter
Our rails
What Gen puts around it.
Tools, models, sessions and fences, each set by a flag.
Tools
The Hanzo MCP toolset rides along on every agent that can take a server for one run.--no-mcpModels
Its model calls go to api.hanzo.ai, metered on your account, for an agent that can name another endpoint. Without the flag, an agent uses the account it is signed in to.--cloud --modelSessions
Signed in, a session is linked to your Hanzo account and streams to mission control as it runs, so it can be followed and picked back up.--resume --no-linkFences
Sessions are approved inside a sandbox; ask before each action instead. A repository’s own .mcp.json stays off until you trust it.--ask --project-mcpgen --claude --cloud # its model calls through api.hanzo.ai
gen --claude --cloud --model enso-ultra
gen --codex --no-link # keep the session on this machine
gen --resume <session-id> # pick a linked session back up
gen --claude -- -c # after --, the agent's own argumentsIn the Hanzo CLI
The same harness, as hanzo dev.
The Hanzo CLI links Gen, so hanzo dev takes the same flags and runs the same session. Use whichever is on your path.
hanzo dev # gen
hanzo dev --claude # gen --claude
hanzo dev --codex --cloud # gen --codex --cloudOpen source
MIT, in the open.
The harness, every agent's launcher and how its output is read are on GitHub, one file an agent.
Questions
Asked plainly.
How do I install it?
Is gen the same as hanzo dev?
Does every agent run on Hanzo’s models?
What does Hanzo bill?
Does Gen change the agent?
What license is it under?
Read
The harness, the agent, the CLI.
Where to get Gen, what it runs, and the same harness in the CLI.