How Forecast AI works.
Forecast AI is a multi-agent forecasting platform for prediction markets and tokenized real-world assets. This documentation describes the live web product, its evidence model, agent system and onchain proof layer.
One system, three forecasting surfaces.
The platform separates collection, specialist analysis, consensus and verification. Users can run Forecast AI's standard seven-agent Swarm or activate a custom standalone agent or custom swarm through Agent Studio.
Forecast Lab
Prediction-market probability forecasts for Polymarket and Kalshi.
RWA Intelligence
Directional RWA Asset Outlooks for Robinhood Stock Tokens.
Agent Studio
User-configured agents for Prediction Markets, RWA Assets or Hybrid (PM+RWA).
Prediction markets are the primary forecast target.
Select a live Polymarket or Kalshi contract, then run the standard Swarm or an active compatible user-created agent. Forecast Lab returns a probability, confidence, individual agent outputs, supporting evidence, citations and market-intelligence context.
Asset outlooks, with prediction markets as supporting signals.
RWA Intelligence starts from a Robinhood Stock Token and its current bid/ask midpoint. The target is whether the reference price may be higher after 24 hours, 7, 30, 90 or 180 days. At the horizon, Forecast AI waits for a verifiable Robinhood quote, compares its midpoint with the original reference, and calculates Brier Scores for consensus and every participating agent. Polymarket and Kalshi events remain supporting signals only.
Build the forecasting identity, then measure it.
Agent Studio supports a custom seven-agent swarm or one standalone specialist. Every identity has a purpose, model configuration, evidence modules, category, RWA horizons and its own proof history.
Custom 7-agent swarm
Configure every specialist separately, then combine all seven results before consensus.
Standalone specialist
Create one purpose-built forecasting agent with a compatible model and evidence stack.
Verifiable before the outcome is known.
Eligible published Prediction Market Forecasts and RWA Asset Outlooks can be committed to Robinhood. The commitment proves that a forecast existed before resolution without treating every private or incomplete run as publishable proof.
Commit
Hash and timestamp on Robinhood.
Resolve
Official market outcome or horizon price.
Score
Brier Score and identity track record.
Evidence keeps its provenance.
Every run can report a provider as used, unavailable, failed or irrelevant. An integration being configured does not mean its evidence is forced into every forecast.
The original forecasting framework remains open source.
Developers can inspect the multi-agent architecture, configure their own model providers, run local forecasts and expose the FastAPI service on their own infrastructure. The hosted product includes additional managed integrations and product workflows.
curl -fsSL https://raw.githubusercontent.com/codebyollie/forecast-agents/main/install.sh | bashCurrent utility is separate from planned utility.
Live now
Currently, a 200,000 $FORAI bond activates one user-created agent identity on Robinhood. Each additional active agent needs its own bond. Unlocking returns the bond and retires the identity while preserving its record. Because Forecast AI is free during testing, holder tiers receive higher usage limits and more analysis and agent runs.
Under discussion
Expanded model access, agent payments, performance rewards and public marketplace economics are potential utilities, not final commitments. They will be discussed further with the community before final specifications are approved.
Live product, active testing.
Forecast Lab, RWA Intelligence, Agent Studio, public result sharing, automatic RWA resolution and the Proof Network are available for testing. The next development phase focuses on deeper profiles, source-correlation controls, public agent discovery and pay-per-use provider routing.
Try Forecast AI