inbrainfun/inbrain

Inbrain is an AI system that automatically creates, monitors, and resolves prediction markets. Prediction markets are places where people place bets on the outcome of future events (like "Will the ETH ETF be approved by Q1 2027?") and the price of a bet reflects how likely participants think the event is. Inbrain's claim is that traditional prediction platforms have no memory or context, so they cannot learn over time. This system adds a memory layer and a set of AI agents that handle the full market lifecycle. Four agents work in sequence. A Signal Agent monitors Twitter, news feeds, on-chain data, and existing prediction platforms to detect trending topics and filter them by engagement speed and influential accounts. A Brain Agent queries a stored knowledge graph to look up how similar past events resolved, then scores each candidate market for verifiability, historical similarity, and community interest, and only creates markets that clear a quality threshold. An Execution Agent monitors active markets in real time, updates the probability estimates as new evidence arrives, and automatically resolves markets when a verifiable outcome can be confirmed. An Analyst Agent publishes daily reports to Discord and Twitter covering the top markets and cases where the AI's probability estimate diverges significantly from what the crowd believes. Every night a process called the Dream Cycle runs a five-phase self-improvement loop: reviewing all past predictions for accuracy, forecasting tomorrow's high-potential markets, finding knowledge gaps in the stored data, updating a calibration model (for example, recording that crypto regulation predictions tend to be overconfident by 8%), and distributing the resulting intelligence reports to the community. The project is written in TypeScript, runs on the Bun runtime, and uses either a local embedded database or PostgreSQL for production. It supports Claude, GPT-4, and Gemini as AI providers via environment variable configuration. Setup takes a single install command and a local initialization step with no container required. The license is MIT.