Polariton

Polariton gives humans and AI agents a persistent place to work together on scientific research. Organize an investigation into connected directions, work with agents on files and analysis, and return to the records behind a result.

A project contains research nodes, each with its own workspace. Workers within a node share its files while keeping separate conversations. Checkpoints preserve file versions you can use to explore another direction. When research needs additional tools, workers can help connect or build reusable capabilities.

Polariton is currently a preview for a single owner on a trusted local machine. It provides research infrastructure; domain-specific tools and experimental workflows build on it. Review generated methods and results before relying on them in research.

Start here

• Install and launch: follow the repository’s macOS or Ubuntu quick start.
• Create your first project: connect a model, give a worker a task, and save a checkpoint.
• Compare two research approaches: use a small synthetic dataset to explore alternatives and inspect their evidence.

For the underlying concepts, see research project structure, capabilities and environments, and research provenance. Check current limitations before using Polariton with existing research data.