Privacy-first local AI
Keep research loops close to your team and avoid turning sensitive product thinking into a third-party data exhaust.
Open source autonomous research desktop lab
Run your own autonomous research lab that never sleeps.
Maabarium helps teams explore, test, and keep better ideas with privacy-first local AI, Git-backed experiments, and a desktop console built for real decision trails.
Local AI
Safe Git experiments
Persistent history
Real desktop console
A visible control centre for autonomous experiments
Shipping in public
Git-backed safety
Explore risky ideas inside isolated workspaces before they touch your main line of development.
Visible decision trails
See what changed, what scored better, and why the winner earned its place instead of relying on memory.
Why teams use Maabarium
From private experiments to visible wins.
The product is built for teams that want more than a chat box. It gives you a repeatable lab for exploring ideas, checking results, and keeping the strongest version without losing the trail.
Privacy-first local AI
Keep your AI workflow private
Run local AI models and keep your research process on your own hardware whenever your setup supports it.
Git-isolated experiments
Test bold ideas without breaking the main branch
Every experiment happens inside a Git-backed workspace so you can compare changes safely before promoting a winner.
Evidence-driven evaluation
Score results instead of guessing
Use structured evaluation so proposals compete on evidence, not vibes, and weak iterations are rejected early.
Persistent history
Track what changed and why it won
Persist experiment history, metrics, proposals, and traces so your team can learn from every run instead of starting over.
Desktop control centre
See the whole loop in one desktop console
Monitor run state, hardware readiness, active blueprints, history, diffs, and setup from a single native interface.
Open source by default
Stay open and adaptable
Maabarium is Apache 2.0 licensed and built in the open, so teams can inspect, extend, and contribute without vendor lock-in.
How the loop works
A keep-winner system instead of endless guesswork.
Maabarium turns autonomous research into a repeatable loop: propose, test, evaluate, and keep the best result. That means faster iteration without losing control.
01
Set a blueprint
Choose the workflow, the models, and the rules you want the lab to follow before the first run starts.
02
Generate proposals
Specialized agents produce competing ideas instead of a single opaque answer, giving you options with different trade-offs.
03
Evaluate in isolation
Run scoring, validation, and sandbox checks inside a contained workspace so changes are tested before they are trusted.
04
Keep the winner
Promote the best result, persist the history, and leave a decision trail the team can inspect later.
Designed for visibility
Know what your lab is doing at every step.
The desktop console keeps the process understandable. You can see live activity, history, hardware readiness, active blueprints, and update state without dropping into ad hoc scripts.
Console
Screenshot from the desktop app
Local metrics and control
See the loop without decoding logs
The console keeps scoring, blueprint context, debate state, and history visible in one place so progress is legible at a glance.
Track experiments safely
Git-backed isolation means the workflow can explore and compare changes without blurring what is safe to keep and what should be rejected.
Keep context attached to each run
Blueprints, runtime signals, evaluation output, and accepted winners stay connected instead of being split across multiple tools.
Quickstart
Install Maabarium with one command.
Start with the one-line installer for the fastest setup on the current supported macOS desktop build. If you prefer a manual path, open the latest signed desktop release instead.
The current hosted desktop release supports Apple Silicon Macs running macOS 11 or later.
macOS Latest release Linux Coming soon Windows Coming soon
curl -fsSL https://downloads.maabarium.com/install.sh | bash Open the latest macOS release. The latest signed macOS release is available if you prefer a manual install path. The generated installer reads the latest updater manifest at runtime, selects the correct architecture, and installs Maabarium into /Applications.
Signed desktop artefacts · latest manifest aware · GitHub Releases
Windows and Linux desktop releases are planned, but the current automated release path is intentionally focused on macOS first.
Open source trust
The lab should be as transparent as the work it produces.
Maabarium is not a closed platform that asks you to trust hidden behaviour. It is an open-source Rust workspace with a desktop console, Git-backed experiment flow, and documentation that teams can inspect directly.
Trust layer
Open source and inspectable
The project is Apache 2.0 licensed, published on GitHub, and designed for teams that want to understand the tool they are depending on.
Trust layer
Local data, clearer boundaries
You decide where the runtime lives, where the data sits, and how far any provider integration should go.
Trust layer
Safer evaluation flow
The workflow favors contained execution, explicit results, and a full trace history instead of silent mutations.
FAQ
Questions teams ask before they adopt a research lab.
The short version: it is open, traceable, local-first in spirit, and built to give experimentation a safer operating model than ad hoc prompting.
Who is Maabarium for?
Teams and solo builders who want a more disciplined way to run AI-assisted research, prompt work, code improvements, or product experiments without losing privacy or traceability.
Does Maabarium only work with remote APIs?
No. The product is designed around a privacy-first direction and can work with local AI setups where your environment supports them, while still allowing remote providers when needed.
What makes it different from using a chat tool and a spreadsheet?
It keeps the whole loop together: proposals, isolated Git workspaces, evaluation, scoring, acceptance, rejection, and persistent history in one system.
Is the project open source?
Yes. Maabarium is open source and released under the Apache 2.0 license, with the repository on GitHub and the user documentation published publicly at docs.maabarium.com.