What it does
Run atlarix in a project folder and ask it to build, fix or explain something. The agent reads the project and plans the work. It edits files and shows a numbered diff for you to approve. It runs shell commands in a per-OS sandbox. You can allow a command once, for the session, or not at all. It also reviews and debugs code. Independent tasks can be handed to sub-agents, and long commands can keep running in the background. To sign in, run atlarix and approve in your browser, or run atlarix login --device on a machine without a browser. NorahLabs makes the CLI.
Capabilities
- Five work modes: Build edits and runs, Explore reads and explains, Plan writes a plan to approve first, Debug chases a failure, and Review reviews changes. Shift+tab switches between them.
- Model choices include Atlarix Auto, which picks a model for each step, and Atlarix Core models. You can also use your own key for providers such as OpenAI, Anthropic, Google, DeepSeek and OpenRouter, or a local model through Ollama or LM Studio.
- Sessions, models, API keys, MCP servers, skills and automations are shared with the Atlarix desktop app.
- atlarix run --prompt "…" does one task and exits, for CI and scripts. The Atlarix cloud agent runs this command inside GitHub Actions.
- Install with a curl script on macOS and Linux, a PowerShell command on Windows, Homebrew or npm. Every method installs the same self-contained program from the Atlarix releases on GitHub, checked against its published checksums.
When to use it
- Your workflow lives in the terminal or in SSH sessions.
- You want work done while you are away, using the Atlarix cloud agent, which runs the CLI in your repository's GitHub Actions and brings back a pull request.
- You want to start a task in the terminal and pick it up in the Atlarix desktop app, or the other way round.
