A team of coding agents in your terminal. One fast Rust binary runs a lead agent that can hand work to specialists running in parallel (scouts, a planner, coders, testers, a reviewer, a verifier and more), with the model of your choice for each, and checks the build before it calls a job done.
› add rate limiting to the API and make sure the tests pass
- Works as a team. Specialists run side by side, each in its own context and optionally its own git worktree. Message or stop any of them while they work.
- Any model, local or cloud. Anthropic, OpenAI, Google, OpenRouter, Groq, DeepSeek, xAI, Mistral, Ollama, LM Studio, llama.cpp, Bedrock, Vertex and Azure, plus Claude and ChatGPT subscription sign-in. Every role takes a fallback chain.
- Goals with a checker. Give it an objective; it keeps going until the project's build and tests pass and an independent verifier agrees.
- Safe by default. Permission modes from read-only planning to full auto, a checkpoint before every
prompt (
/rewind), and agents nevergit push. - Sees and acts. Web search and fetch, a real Chromium browser, screen capture and input for playtesting apps and games, language-server diagnostics after every edit.
- Remembers. Per-project memory and a searchable code index that carry across sessions.
- Makes assets. Images, video, 3D models and audio, and sprite sheets packed for Godot, Unity, Roblox, Unreal or plain JSON.
- Extensible. MCP servers, hooks on any tool call, Luau extensions, custom agents, skills, slash commands, packages, and an embeddable Rust API.
Linux and macOS:
curl -fsSL https://raw.githubusercontent.com/sinnafuls/azazel-cli/main/install/install.sh | shWindows (PowerShell):
irm https://raw.githubusercontent.com/sinnafuls/azazel-cli/main/install/install.ps1 | iexThen open a new terminal, cd into a project and run azazel. The first start walks you through signing in
or adding an API key. Update later with azazel update.
From source (Rust 1.99): clone the repository and run ./install/install.sh or .\install\install.ps1.
azazel # interactive UI in the current project
azazel -p "explain src/main.rs" # one prompt, answer on stdout
azazel goal "make cargo test pass" # work until the checker verifies it
azazel bg "fix the flaky test" # in the background; azazel attach <id> opens it later
azazel schedule add every 1h "triage new issues" # on a clock, or --on ci-failure / pr-comment / slack:<channel> / file:<glob>
azazel --ssh build-box # the same UI on a session that runs on another machine
azazel doctor # what is configured and what is missing
azazel migrate --dry-run # what would come over from pi, omp, jcode, Claude Code, Codex CLIStart with getting started; every feature has its own page in docs/. To work on azazel itself, read AGENTS.md.
Found a bug or want something added? Open an issue
with the bug report or feature request template. Search the existing issues first, keep one problem per issue, and
never paste API keys, tokens or auth.json; security problems go to a
private report. For a bug, include
azazel --version, your OS and terminal, the steps, and a screenshot if anything is drawn wrong.
Pull requests are welcome: CONTRIBUTING.md has the ground rules and the checks to run.
Licensed under either of Apache-2.0 or MIT, at your option.