# LabRat

An autonomous multi-agent research orchestrator that runs specialized AI agents for literature mining, data synthesis and hypothesis testing, allocating shared compute as tradeable market credits.

A researcher defines a hypothesis and lets a colony of role-specific agents explore it: agents earn lab credits for high-value findings and spend those credits on compute, API calls or memory priority, with an orchestrator and a market-allocation engine mediating who gets resources next. Findings accumulate in a shared memory pool that feeds a final experiment report, which can call out to the Claude API for a summary and the OpenAI API for a critique.

Agents can read papers in more than 32 languages and synthesize findings in a chosen output language, research loops continue until statistical saturation or the budget runs out, and every run produces a deterministic reproduction bundle containing its configuration, agent logs and credit ledger. Activity can be watched from a terminal or a lightweight web dashboard, and the whole colony is configured from a single YAML file.

Website: https://github.com/Araaf7/swarm-factory