Release workflow already builds dev-* binaries. Update npm packages (dev-cli, hanzo-node) and preview-build workflow to reference dev-* directly, eliminating the download-and-rename indirection. Removed stale dev-cli/.pack directory. Cleaned code-* assets from v0.6.83-v0.6.85 GitHub releases.
Hanzo Dev
Hanzo Dev (@hanzo/dev) is a fast, native coding agent for your terminal. Built in Rust on top of openai/codex, it adds multi-agent orchestration, CLI agent piping, browser automation, a rich theme system, and a full chat TUI.
Install
npm install -g @hanzo/dev
dev
Or run directly:
npx -y @hanzo/dev
The binary is named dev. The package also installs coder as an alias.
Key Features
- Multi-agent orchestration —
/plan,/solve,/codecoordinate Claude, Gemini, Qwen, and GPT - CLI agent piping — spawns
claude,gemini,qwenas sub-processes with output streamed into the TUI - Auto Drive — autonomous multi-step task execution with self-healing
- Auto Review — background ghost-commit watcher for continuous code review
- Rich TUI — Zen mode, 20+ themes, streaming markdown, agent terminal (
Ctrl+A), session management - Browser integration — CDP support, headless browsing, inline screenshots
- MCP support — Model Context Protocol with hot-reload and 260+ tools
- Skills — dynamic tool injection with live reload
- Upstream sync — automated merge from
openai/codexevery 30 minutes
Authentication
- ChatGPT sign-in (Plus/Pro/Team) — run
dev, pick "Sign in with ChatGPT" - API key —
export OPENAI_API_KEY=sk-... && dev - Device code — headless environments auto-prompt device code flow
CLI Reference
dev [options] [prompt]
Options:
--model <name> Override model (e.g. gpt-5.1, claude-opus-4-6)
--read-only Prevent file modifications
--no-approval Skip approval prompts
--config <key=val> Override config values
--oss Use local models (Ollama)
--sandbox <mode> Sandbox level (read-only, workspace-write)
--url <endpoint> Custom API endpoint
--debug Log API requests/responses
--help Show help
--version Show version
Configuration
Config: ~/.hanzo/config.toml (reads ~/.code/ and ~/.codex/ for compatibility)
model = "gpt-5.3-codex"
model_provider = "openai"
approval_policy = "on-request"
model_reasoning_effort = "medium"
[tui.theme]
name = "dark-zen"
zen = true
[mcp_servers.filesystem]
command = "npx"
args = ["-y", "@modelcontextprotocol/server-filesystem", "/path"]
| Variable | Purpose |
|---|---|
HANZO_HOME |
Config directory (default: ~/.hanzo) |
OPENAI_API_KEY |
OpenAI API key |
ANTHROPIC_API_KEY |
Anthropic API key |
OPENAI_BASE_URL |
Custom OpenAI-compatible endpoint |
Agent Piping
Install companion CLIs to unlock multi-agent orchestration:
npm install -g @anthropic-ai/claude-code @google/gemini-cli @qwen-code/qwen-code
Then use /plan, /solve, or /code — Hanzo Dev detects installed CLIs and orchestrates them in parallel git worktrees.
Custom agents: create YAML-frontmatter .md files in ~/.hanzo/agents/ and invoke with /use <name>.
License
Apache 2.0 — community fork of openai/codex. Not affiliated with OpenAI.
See the main README for full documentation.