What Maabarium Is

Maabarium is an open-source autonomous research lab built around a simple loop:

1. Propose a change.
2. Apply it in an isolated Git-backed workspace.
3. Evaluate the result.
4. Keep the winner or reject it.
5. Persist the traces, logs, scores, and outcomes.

That loop matters because it keeps experimentation disciplined. You are not just prompting a model and hoping for the best. You are operating a system that records what changed, how it was evaluated, and what survived.

Who this is for

Maabarium works best for:

• builders iterating on prompts, adapters, code, and research workflows
• teams who want a visible record of why a result was kept
• people who prefer local-first tooling and Git-native workflows

What the desktop app gives you

The desktop app is the easiest way to operate Maabarium day to day. It gives you:

• live workflow controls
• visible telemetry and run analytics
• workflow library browsing
• readiness checks, Git setup guidance, and maintenance actions
• retained winner scoring and persisted history in one place
• persisted history and logs in one place

What power users still get

This is not a simplified wrapper that hides the internals. You still have direct access to:

• blueprint files
• CLI execution
• persisted runtime files
• logs and experiment history
• Git workspaces and branches created during runs

If you care about optimised local builds, runtime performance, and advanced platform-specific behaviour, continue into Build Profiles and Runtime Behaviour.

If you want the deeper system details, continue into Blueprint Reference and CLI and Runtime.