Forecast AI

Open-source multi-agent AI for prediction markets

Artificial Intelligence

GitHub

Tech

Web3

Hey Product Hunt 👋 I’m the developer behind Forecast AI. I started building Forecast AI because I kept coming back to the same problem with prediction markets: there is an enormous amount of information available, but turning all of it into one useful decision is still surprisingly difficult. A market can depend on breaking news, deeper research, macro data, market structure, onchain activity, social sentiment, and community discussions - all at the same time. Asking a single AI model to understand everything and return one probability never felt like the right approach. So I started experimenting with a different idea: instead of one AI trying to do everything, what if multiple specialized agents independently looked at the same market from different perspectives? That became Forecast AI. Today, Forecast AI runs a swarm of 7 specialized agents: 📰 News Agent - follows current news and events 🔬 Research Agent - performs deeper evidence-based research 📊 Macro Agent - looks at economic and macro signals ⛓️ Onchain Agent - analyzes relevant blockchain activity 📈 Market Agent - studies market probabilities and structure 💬 Social Agent - looks at broader social sentiment 🤖 Reddit Agent - analyzes community discussions and retail sentiment For deeper research, we’ve also integrated FactsAI by Polyfactual, giving our research layer access to source-backed research and citations. The important part is what happens next. These agents don’t simply produce seven separate answers. Their findings are brought into our consensus layer, where different signals are weighted, disagreements between agents are considered, and everything is combined into a final probability forecast with supporting evidence, reasoning, and a confidence score. The goal isn’t to build another AI chatbot that gives you a YES or NO. I want Forecast AI to become an intelligence layer for prediction markets -something that helps people understand why a market may be mispriced before they make their own decision. Forecast AI currently works with markets across Polymarket and Kalshi, including markets mirrored through Robinhood Predict. The final output can also be handed off to Robinhood’s Agentic Trading MCP, while the user remains in control of execution. Another decision I made early was to keep the core infrastructure open source. Developers can inspect the agents, understand how the system reaches a forecast, modify it, add their own data sources, and run the entire swarm themselves. I think transparency matters especially when AI is being used to support financial and prediction-market decisions. But self-hosting shouldn’t be required forever. The next major thing I’m building is bringing the entire agent workflow directly to our website: Choose a market → launch the 7-agent swarm → research → consensus → final forecast. No installation and no local environment. There will be a free level so anyone can use Forecast AI, while $FORAI holders will progressively unlock deeper analysis, stronger capabilities, and additional features. So you can choose what better works for you and you will be able to test everything directly on our environment before you decide to run it on your own. We’re still very early. I’m shipping quickly, learning from users, and there is a lot I want to improve. If you try Forecast AI, I’d especially love to hear what you think about the agent approach, what prediction markets you’d like us to support next, and what you’d want to see in the on-site swarm experience. Thanks for checking out Forecast AI and being part of the beginning. Back to building.