zeekayandHanzo Dev
23a7fe1109
Branding. The reported "OpenAI Codex" was not a banner — the TUI banner has said "Welcome to Code" all along. It was the model describing itself. Every request carries BASE_INSTRUCTIONS, and `zen5-coder` matches none of the OpenAI slug prefixes in find_family_for_model, so it lands on derive_default_model_family -> core/prompt.md, whose first line read "You are a coding agent running in the Codex CLI ... led by OpenAI". The agent was told who it was, and it answered honestly. Fixed the harness identity in all eight prompts (keeping each model's own identity: "based on GPT-5" stays, the harness around it becomes Hanzo Dev), the per-turn header, approval and model pickers, the login success page, MCP tool titles and approval prompts on both client and server sides, seatbelt policy header, cloud copy, two stale User-Agents, and the CLI hints that named the wrong binary. Golden JSON fixtures moved with the code they assert. Deliberately NOT changed, because they are wire and not text: `OPENAI_*` and `CODEX_*` env vars, the `codex`/`codex-reply` MCP tool names and the `code-mcp-server` server name (hosts key allowlists off them), `codex.*` OTEL event names, the `openai` provider id, model slugs, `~/.code` on disk, and the `https://api.openai.com/auth` JWT claim namespace. Control. `cli/src/proto.rs` was a complete bidirectional control plane — Submission JSON lines in, Event JSON lines out, the whole `Op` set including QueueUserInput — exported from lib.rs with no subcommand to reach it. Wired it up as `dev proto`. That is the difference between interrupting a turn and steering one: `app-server`'s sendUserMessage hardcodes Op::UserInput, which calls sess.abort() and restarts the turn. $ printf '%s\n' '{"id":"s1","op":{"type":"interrupt"}}' \ '{"id":"s2","op":{"type":"shutdown"}}' | dev proto {"id":"","event_seq":0,"msg":{"type":"session_configured", "session_id":"4d82071e-...","model":"zen5-coder",...}} {"id":"s2","event_seq":1,"msg":{"type":"shutdown_complete"}} Also: upstream-merge.yml polled openai/codex every 30 minutes. That is the wrong upstream (codex-rs reaches us through just-every) and a cron cannot work against a disjoint history anyway — it needs a hand-supplied merge base. Retargeted to just-every/code and made it manual, pointing at the procedure in LLM.md. Purged @just-every/code from the release and recovery pipelines, which would otherwise have published our binaries under their scope. Co-authored-by: Hanzo Dev <[email protected]>
Hanzo Dev
Hanzo Dev is our coding agent. It runs in your terminal, reads and edits the
repository you point it at, runs commands in a sandbox, and drives a browser
when a task needs one. Models are served by Hanzo AI —
enso and the zen5 family — over the metered gateway at api.hanzo.ai.
It ships inside the Hanzo CLI. hanzo dev is the command.
 A session in progress |
 Reviewing a diff |
|---|---|
 Driving a browser |
 Several agents on one task |
Install
curl -fsSL https://hanzo.sh | sh
hanzo auth login
hanzo auth login signs you in through Hanzo IAM and links the machine to your
account. Then, from any repository:
hanzo dev
Give it the task on the command line to run headless instead:
hanzo dev "fix the failing tests in ./core and explain what was wrong"
Choosing a model
Model ids come from the gateway, which is the only authority on the catalog — there is no allowlist in the client.
hanzo dev --model enso # our frontier model
hanzo dev --model enso-ultra # the largest reasoning tier
hanzo dev --model zen5-coder # tuned for code
Browse everything available with curl https://catalog.hanzo.ai/v1/models.
Backends
Hanzo Dev is the default and the one we build. The same session launcher can also drive two other agents if you already have them installed, so you can keep one set of habits across all three:
hanzo dev # our agent (default)
hanzo dev claude # or: hanzo code --claude
hanzo dev codex # or: hanzo code --codex
hanzo dev is a spelling of hanzo code dev, not a second implementation —
every spelling resolves through the same launcher.
Options
hanzo dev [OPTIONS] [BACKEND|TASK] [TASK] [-- <PASSTHROUGH>...]
--model <MODEL> Gateway model id, e.g. enso, enso-ultra, zen5-coder
--backend <NAME> dev (default) | claude | codex
--brand <BRAND> Tenant for auth: hanzo | lux | zoo | pars | bootnode
--ask Ask before each action instead of auto-approving (alias: --safe)
--no-sandbox Drop the sandbox and auto-approve. Per-invocation, never persisted
--no-route Call the backend's own model account instead of the metered gateway
--no-mcp Do not attach the Hanzo MCP toolset
--project-mcp Also load the repo's own .mcp.json servers (off by default — a repo is untrusted)
--link / --no-link Stream the session to Hanzo cloud, or never
--resume <ID> Resume a prior linked session by its cloud session id
-c, --config <FILE> Use a specific config file
-v, --verbose Increase logging verbosity
Anything after -- is passed verbatim to the backend.
Two defaults worth knowing. Sessions run sandboxed and auto-approve inside that
sandbox; --ask tightens it, --no-sandbox removes it. Model calls route
through api.hanzo.ai so usage is metered on your Hanzo account; --no-route
sends them to the backend's own account instead.
In a session
Type / for the full list. The ones you will reach for:
/plan create a comprehensive plan (multiple agents)
/solve solve a challenging problem (multiple agents)
/code perform a coding task (multiple agents)
/auto work autonomously on long tasks with Auto Drive
/agents create and configure agents
/browser open the internal browser
/chrome connect to a running Chrome over CDP
/diff show the git diff, including untracked files
/review review your changes for potential issues
/undo restore the workspace to the last snapshot
/branch work on a branch in a worktree
/merge merge the worktree branch back to the default
/model choose model and reasoning effort
/reasoning change reasoning effort (minimal/low/medium/high)
/theme switch color themes
/status session configuration and token usage
/limits weekly and hourly rate limits
/mcp manage MCP servers (status/on/off/add)
/init write an AGENTS.md for this repository
/resume resume a past session for this folder
/new start a new chat
Full list with descriptions: docs/slash-commands.md.
Project context
Run /init in a repository and Hanzo Dev writes an AGENTS.md describing what
it found. Edit that file — it is read at the start of every session in that
directory, so it is where project conventions, architecture notes and "do not
touch this" rules belong.
Tools
The Hanzo MCP toolset is attached by default: filesystem, shell, browser, and
the rest of the Hanzo surface. Add your own servers with /mcp add, or per
repository in .mcp.json — the latter is off unless you pass --project-mcp,
because a repository you just cloned should not get to run a server with your
session's key.
Documentation
Building from source
git clone https://github.com/hanzoai/dev.git
cd dev
./build-fast.sh
./build-fast.sh is the required check — it must pass cleanly, warnings
included. See DEVELOPING.md and LLM.md for the
workspace layout and the rules that apply inside it.
Contributing
Branch, make the change, make ./build-fast.sh pass, open a pull request.
Bugs and feature requests: github.com/hanzoai/dev/issues.
License
Apache-2.0 — see LICENSE.
Lineage
Hanzo Dev is a fork of just-every/code, which is
itself a fork of openai/codex. Both are Apache-2.0,
and this repository preserves the upstream LICENSE and NOTICE. Hanzo Dev is not
affiliated with, sponsored by, or endorsed by either project.