Documentation

Getting started

Install Orbit or the Loom CLI, connect Galaxy, and run your first agentic analysis.

Getting started

Orbit and Loom turn a working directory into a co-scientist project. You chat with an agent; it plans, runs work on Galaxy or locally, and records everything in a durable notebook.md. This page gets you from zero to your first analysis.

1. Install

Pick whichever shell you prefer — they run the same brain.

Orbit (desktop app)

Download an installer from the Releases page. Each build bundles its own Node, uv, and Loom, so there is nothing else to set up. The macOS build is signed and notarized.

  • macOS — open the .dmg and drag Orbit to Applications.
  • Linux — install the .deb (sudo dpkg -i orbit_*_amd64.deb && sudo apt-get install -f) or .rpm, then run orbit.
  • Windows — remote-only (no local shell). Currently ships as an unsigned .zip — unzip and run orbit; a signed installer is on the way.

Loom (CLI)

The CLI needs Node 22.19+ and, for Galaxy MCP, uv.

npm install -g @galaxyproject/loom
loom

Or run it once without installing:

npx @galaxyproject/loom

2. Connect Galaxy

With Galaxy connected, the agent can survey the workflow registry and tool catalog while drafting plans, and route individual steps to Galaxy. Provide a Galaxy server URL and an API key (Orbit stores credentials in your OS keychain; the CLI reads GALAXY_URL / GALAXY_API_KEY).

Galaxy access flows through Galaxy MCP — the bridge that lets any agent search and run tools, drive workflows, and manage histories.

Galaxy work is optional. Loom is useful for local exploration and planning even before you connect a server.

3. Run your first analysis

  1. Open a directory. Start Orbit (or run loom) in an analysis folder. A notebook.md is created on first launch and committed to git.
  2. Chat. Ask questions, drop file paths, request data lookups. None of this requires a “plan” — the conversation is just a conversation.
  3. Ask for a plan. The agent drafts it in chat as a markdown section and waits for your approval. Then it shows you the parameters and waits again.
  4. Watch it run. Once approved, the plan is written into notebook.md and executed. Galaxy steps are tracked with live loom-invocation blocks; the Activity tab streams tool calls and job status.
  5. Come back tomorrow. Open the same directory and the notebook is the project. Sessions resume automatically.

Where the record lives

Everything the agent decides and does is written to notebook.md as plain markdown, and auto-committed to git. There is no parallel hidden state store — the notebook is the durable working record, and Galaxy histories remain the computational truth. See Concepts for the full model.