Quick start
Requires Rust 1.88.0+ and Cargo, network access, and a supported provider account or API key.
Connect a provider from Mission Control:
/login
Choose Codex browser login, a Claude subscription via installed Claude Code CLI, an Anthropic API key, or a configured OpenAI-compatible provider. Then ask a question or describe a change. To queue the first prompt at launch:
Use magi-code --help for CLI options or /help in Mission Control. See Getting started for provider setup and source installs.
A coding harness you can inspect
magi-code keeps conversations, tool activity, and code changes together in one terminal workspace. Its compact core prompt, coding tools, and configurable agents let you shape the workflow without hiding how work gets done.
- Inspect: Follow tool calls in the transcript and Activity Tree, including submitted parameters, status, and captured output.
- Direct: Choose models and agent profiles, add repository instructions and skills, or delegate independent work to parallel subagents.
- Review: Inspect current worktree diffs, leave line comments, and send unresolved feedback back to the agent.
- Own: Sessions, settings, checkpoints, and agent profiles live on your machine by default.
Built for ongoing work
- Resume saved conversations, compact long sessions, or rewind conversation history and captured file-tool changes.
- Search and edit code with focused file, text, and AST-aware tools; run builds and checks through the shell; use Code Mode, enabled by default, for JavaScript tool batches with the same approvals and records.
- Connect MCP tools, optional language-server diagnostics, public web research, and image inspection.
- Use Codex, Anthropic, or custom OpenAI-compatible providers, with separate model choices for primary agents and specialists.
- Enable optional Jev-native functionality for skill suggestions, command and prose checks, prompt-injection protection, completion verification, and inferred Bash file tracking.
Documentation
Start with the website for an overview and the user guide for everyday workflows.
| I want to... | Read |
|---|---|
| Get running | Getting started |
| Learn Mission Control | Mission Control |
| Configure providers or settings | Providers and models, Configuration |
| Manage sessions and context | Sessions and context |
| Customize or delegate agents | Customization, Agents and delegation |
| Look up exact behavior and limits | Technical reference |
Safety
magi-code can edit files and run commands. Shell guardrails are not an OS-level sandbox; isolate untrusted repositories separately.
Runtime data defaults to ~/.magi-code; set MC_HOME to use another location. Inference sends prompts, selected repository context, and tool results to your configured provider. Keep credentials out of settings.json and never commit auth files or secrets. See Security notes.
Development
See Contributing, verification, profiling benchmarks, and the release checklist. Historical ADRs, PRDs, and plans live in the documentation archive.
License
Magi-code uses Apache License 2.0 with Commons Clause. It is source-available, not OSI-approved open source or plain Apache-2.0.
You may modify it, redistribute it for free, and use it commercially inside your organization, subject to the required notices. You may not sell magi-code itself or a product or service whose value comes entirely or substantially from its functionality. See LICENSE for the full terms.