Forecast AI
Seven specialized agents feed a Consensus Engine that turns market signals into calibrated probability forecasts for Kalshi and Polymarket traders.
Forecast AI Multi Agent Consensus Bot for Polymarket and Kalshi
What is Forecast AI?
Forecast AI is an open source, seven agent forecasting system that turns Polymarket, Kalshi, and Robinhood Predict data into calibrated probability forecasts and trade recommendations for prediction market traders.
Ready to try Forecast AI?
Open the official site and get started in a few clicks.
Forecast AI Overview
Forecast AI is an open source multi agent system that turns raw market signals into calibrated probability forecasts for prediction markets. Instead of asking one language model for a guess, it runs seven specialized agents in parallel, each one working a separate data stream: news, social sentiment, community discussion, structured research, macro indicators, on chain activity, and live order book data from Kalshi and Polymarket. Each agent produces its own probability estimate, and a Consensus Engine combines all seven into a single calibrated number with a confidence score and a short trade recommendation.
The tool is built for people who want an evidence based edge on event outcomes. Active traders on Robinhood Predict use it to check their own read against a structured forecast before placing a trade. Developers who want a self hostable intelligence layer can clone the code and extend it with their own agents or LLM providers. Anyone watching Kalshi event contracts or cross checking Polymarket pricing can use the public feed for a second opinion. It does not place trades for you. It formats a recommendation you can review and act on yourself through Robinhood's Agentic Trading MCP endpoint.
The core idea behind the design is that a single model prompt tends to lean on recent headlines, miss context, and struggle when evidence points in different directions. Forecast AI splits the work instead. Each agent stays inside its own domain, and the Consensus Engine applies weights based on each agent's past accuracy, then runs a calibration step that pulls uncertain forecasts back toward 50 percent rather than letting one noisy signal dominate. A Historical Memory layer stores past forecasts and outcomes so the weighting improves as more markets resolve. The project's own benchmarks report a 24 percent improvement in Brier score over a plain GPT 4 baseline under matched conditions.
What sets it apart from a typical trading bot or single prompt wrapper is the separation between analysis and execution. The system hands off a structured recommendation, action, target contract, confidence level, and risk notes, and leaves the actual buy or sell decision inside the user's own Robinhood account. The full agent architecture, prompts, and consensus math sit in the open repository, so the reasoning behind any given probability is not hidden behind a closed API. A public dashboard also shows the live consensus next to real market prices across Kalshi, Polymarket, and Robinhood Predict, so you can see how the model performs before trusting it with real capital.
Forecast AI Key features
-
Seven Specialized Agents
Each agent works on one data stream: news, social, Reddit, research, macro, on-chain, and market data. They run in parallel and never mix their inputs.
-
Consensus Engine
This engine combines all seven agent outputs into one probability. It weighs each agent by past accuracy and pulls low confidence forecasts toward 50 percent.
-
Historical Memory
Past forecasts and outcomes get stored so the system learns which agents to trust more over time. This helps it stay accurate on markets that run for months.
-
Robinhood Trade Recommendations
The consensus turns into a structured prompt with an action, target contract, and confidence level. You paste this into Robinhood's Agentic Trading MCP and decide whether to trade.
-
Public Read Only API
Anyone can pull live consensus data for crypto, macro, and tech topics. Refresh times run from two hours up to twelve hours depending on the market.
-
Multiple LLM Providers
The system uses OpenAI by default and switches to Gemini if that fails. It also supports Anthropic, OpenRouter, and local Ollama models.
-
CLI and Server Modes
Developers run one shot forecasts, watch loops, or a full FastAPI server from the terminal. This makes it easy to fit into other tools or pipelines.
-
Open Source Codebase
The full code sits on GitHub under Apache 2.0. Nobody's funds or credentials pass through the system since it only formats trade suggestions.
-
$FORAI Holder Tiers
Holding the token unlocks higher analysis limits, deeper agent runs, and stronger models on the hosted version. The core software stays free and open for anyone to self host.
-
Cross Venue Market Data
The Market Agent reads live prices from Kalshi and Polymarket to ground each forecast in real pricing. This keeps the consensus tied to what traders are actually paying.
Forecast AI Fees
-
Core software The open source core has no license fee and can be self hosted at your own compute and LLM API cost.
-
FactsAI research Optional FactsAI deep research requests cost 0.012 dollars each when enabled.
-
Public API The public API is free but rate limited and capped by a server side monthly compute budget guard.
-
Hosted tiers Free accounts get limited analyses, while higher analysis volume and stronger models unlock at 1,000 and 10,000 FORAI token holdings.
-
Trading fees There are no platform trading fees since Forecast AI only recommends trades and never executes them.
How to use Forecast AI
-
Clone the repository
Copy the codebase from the official GitHub repository.
-
Run the setup command
Execute the interactive setup script to configure your LLM API keys and environment variables. You can also add an optional FactsAI key for source-backed research.
-
Start the monitoring loop
Launch the public feed or a continuous watch loop using the command line interface. This begins background tracking.
-
Generate a forecast
Provide a market query or identifier to generate a one-shot prediction.
-
Review the consensus report
Read the final probability, confidence score, and agent reasoning. The system filters anomalies and applies calibration to the data.
-
Generate the trade recommendation
Convert the consensus output into a structured trade prompt for Robinhood.
-
Execute or discard the trade
Paste the prompt into an authenticated client connected to the Robinhood Agentic Trading MCP. You must manually approve or reject the action inside your own account.
Forecast AI Screenshots
Forecast AI Review
Pros
- It runs seven specialized agents in parallel so no single data stream skews the forecast.
- The Consensus Engine applies Bayesian averaging and historical accuracy weights before giving a final probability.
- Public benchmarks show a 24 percent improvement in Brier score over a vanilla GPT-4 baseline.
- Every forecast includes a short rationale from each agent, so users see the reasoning trail.
- It never places trades itself, so users keep full control of execution inside their own Robinhood account.
- The full codebase is open source under Apache 2.0, so anyone can audit or extend it.
- Users can self host the system and keep their own API keys and compute under their own control.
- It supports OpenAI, Gemini, Anthropic, OpenRouter, or local Ollama, and switches providers automatically if one fails.
- The Market Agent pulls live order books and midpoints from both Kalshi and Polymarket to ground forecasts in real prices.
- Historical Memory stores past forecasts and outcomes, so agent weightings improve over time.
- The CLI and FastAPI server let developers run one shot predictions or continuous watch loops from the terminal.
- The Robinhood Recommendation Engine formats a ready to paste prompt with action, target contract, and risk notes.
- The public read only API gives free access to live consensus data for crypto, macro, and tech topics.
- A rate limit and monthly compute budget guard keep the public feed stable and free from abuse.
- The live dashboard at forai.tech shows real time consensus against market midpoints across three venues.
- The sole official $FORAI contract address is published on the website, GitHub, and X for easy verification.
- The system holds no custody of user funds or credentials at any point in the process.
Cons
- Users must supply and pay for their own LLM API keys to run the agents.
- The tool never places trades, so users must manually copy each recommendation into Robinhood's MCP client.
- Public API refresh cycles run two to twelve hours, too slow for fast moving markets.
- The public feed enforces rate limits and a monthly compute budget cap.
- Optional FactsAI research queries add a per call cost on top of normal use.
- Higher analysis limits and stronger models are locked behind $FORAI token holding tiers.
- Local setup requires cloning the repo and configuring multiple environment variables through the CLI.
- The project is still in active development, so parts of the pipeline may change or break.
- Coverage is limited to a curated set of topics like crypto, macro, and tech.
- Users must independently verify the $FORAI contract address since no official custody or verification step is built in.
Our verdict
Forecast AI is an open source tool that runs seven agents (news, social, Reddit, research, macro, on chain, and market data) in parallel, then blends their outputs into one calibrated probability with a confidence score and a trade suggestion. The Consensus Engine is the strongest part of the design, since it weighs each agent by past accuracy and pulls weak signals back toward 50 percent instead of letting one loud data source skew the call. The project claims a 24 percent improvement in Brier score over plain GPT 4, and while that number comes from the team itself, the fact that every agent's reasoning trail is visible makes the claim easier to check than most. It falls short in one clear way: it does not place trades, so the output is only as useful as the person reading it and deciding what to do inside their own Robinhood or Polymarket account. This tool fits developers who want a self hosted forecasting layer and active traders who want a second opinion beyond the order book, but anyone looking for a bot that trades on autopilot should keep looking elsewhere.
Is Forecast AI safe & legit?
codebyollie built Forecast AI as an open source project under the Apache 2.0 license, so anyone can read the full code on GitHub. It never touches your funds or logs into your accounts; it only produces a trade recommendation that you must manually approve inside your own Robinhood session, and the team posts updates and the one official $FORAI contract address on their X account and site. No major red flags stand out, but the project is still early stage, so check the code yourself and confirm the contract address before touching the token.
X account intel @forecast_agents
-
Based in Eastern Europe (Non-EU)
-
Joined X October 2025 10 months ago
-
Username changes 7 renames last on Jun 30, 2026
-
Signup source Eastern Europe (Non-EU) App Store
Public profile signals from X. Use as one input alongside other due-diligence.
Wallet blacklist scan checked Aug 15, 2026
- MetaMask Not flagged
- Phantom Not flagged
- ScamSniffer Not flagged
- EtherAddressLookup Not flagged
- Keplr Wallet Not flagged
Domain forai.tech checked against public crypto wallet blacklists.
Forecast AI Updates
-
GM, Forecasters.
New week has started. What we shipped last week. Quick Forecast AI recap:
• Launched the Forecast AI Proof Network on @RobinhoodApp Chain Mainnet, allowing agent forecasts to be committed onchain before markets resolve
• Joined the @Polyfactual team for a podcast to discuss prediction markets, AI agents and what we’re building at Forecast AI
• Announced our partnership with @bravadotrade and integrated its Trader Analytics API, adding ranked trader positions, performance data and Smart Money confirmation to our Swarm analysis
• Shared the first look at Agent Studio - our upcoming feature for creating custom seven-agent swarms or standalone agents with configurable models, intelligence providers and private APIs
• Improved agents' performance, citations and source provenance
This week:
• Proof resolution, Brier Scores and public agent rankings
• Releasing Agent Studio
• Expanded $FORAI utility around agent activation and participation
• More intelligence integrations and product improvements
Let’s keep building.
- 3 replies
- 1 reposts
- 11 likes
- 238 views
-
We’re excited to announce our partnership with @bravadotrade.
We’ve integrated Bravado’s Trader Analytics API, giving Forecast AI access to ranked non-bot Polymarket traders, their active positions, win rates and PnL.
Bravado now powers an additional smart-money layer for our seven-agent swarm - showing whether top trader positioning confirms or diverges from the Swarm forecast and helping surface additional market risks.
- 8 replies
- 5 reposts
- 17 likes
- 1.1K views
-
GM, forecasters.
Here’s a first look at the upcoming Forecast AI Agent Studio.
https://forai.tech/profile/agents
Check out the preview and tell know what you think. Feedback/suggestion is always welcome.
You’ll be able to create your own agents, configure their models and intelligence providers, connect your own APIs, and build verifiable track records over time.
The design and functionality are still evolving, with more options coming before launch.
The staking model is also being refined, but the core idea remains: stake $FORAI to activate an agent. Unstaking retires the agent, meaning it can no longer participate in network activities, rewards, or agent payments.
Its previous forecasts, scores, and reputation remain permanently verifiable.
P.S. I’ll keep working on my video-making skills 😅- 4 replies
- 2 reposts
- 16 likes
- 642 views
-
GM, Forecasters. Happy weekend.
Take some time to disconnect, touch grass and enjoy the day. Sometimes stepping away is the best way to come back stronger.
We’ll keep you updated from our side.
Have a great weekend.
- 2 replies
- 1 reposts
- 12 likes
- 454 views
-
GM, Forecasters.
Finally finished the article.
https://x.com/forecast_agents/status/2088273317639622942
It’s a little long, but it covers what Forecast AI is, how it works and what’s coming next.
Also preparing a batch of videos for you.
In parallel, finishing one big integration and working on the new features we already have planned. Will also share the completed roadmap.
Sorry, I’ve been a little slower these last few days and less active on x, but I'm back.
Have a wonderful day. Stay tuned.- 5 replies
- 4 reposts
- 16 likes
- 760 views
-
GM, Forecasters. Sorry for being a little less active here lately. Quick update:
- Currently working on a new data provider integration - already in testing. Announcement soon.
- Improving what's already live.
- Finishing the article and preparing a proper demo video to show how everything works under the hood.
- Working on the next steps for token utility.
One more target for the end of the week: let you create and run your own forecast agents. Run your forecasts at https://forai.tech and share your feedback.
Stay tuned- 8 replies
- 1 reposts
- 17 likes
- 783 views
-
HM, Forecasters. Have a great and productive day.
Keep cooking. I'll need some help from the community soon. Stay tuned.
- 2 replies
- 2 reposts
- 11 likes
- 930 views
-
Huge thanks to the @basedlourie for having us and to everyone who joined and listened.
Podcast record will be live shortly and will be shared on our channels.
Criticism, questions and feedback are always welcome.
It was my first time doing something like this, and definitely a great experience. I know I need to do more of these.
We are starting our own community calls next week, so I'll have more opportunities to talk directly with you, share what we're building, and hear your feedback. Anyone is welcome to join. I'll drop the first date and time soon.
Jump into our Telegram - I'm always happy to chat with everyone. Your feedback is valuable.
Back to building.- 4 replies
- 3 reposts
- 12 likes
- 1.9K views
-
Our Proof Network is now live on @RobinhoodApp mainnet. One more step toward a verifiable onchain reputation layer for AI agents.
How it works: Before a market resolves, forecasts from our 7-agent Swarm are timestamped and committed onchain - making them verifiable and preventing changes after the fact.
Once the outcome is known, each agent receives a Brier Score based on its actual forecasting accuracy.
This enables:
• Public, verifiable agent rankings and transparent forecasting track records
• User-created agents with custom models & data providers - already under development. You'll be able to create and run your own agents
• A future agents marketplace based on proven performance
• Expanded $FORAI utility for creating, running and publishing agents, accessing advanced intelligence, and participating in the network - under development.
Now main focus is on resolution testing, more LLM models and data providers integrations, public agent profiles, user-created agents, and expanded $FORAI utility.
Stay tunes. We've just started...
- 4 replies
- 3 reposts
- 9 likes
- 875 views
-
HM, Forecasters.
Have a great and productive day.
We’ve got some news and updates to share today.
More building, more shipping, more coming.
Stay tuned.
- 5 replies
- 2 reposts
- 14 likes
- 955 views
-
Third time's the charm.
Sorry for keeping you waiting. Our Podcast with @Polyfactual is locked in.
Tuesday 🕖 7 PM EST
Looking forward to seeing everyone there.
- 5 replies
- 3 reposts
- 18 likes
- 1.4K views
-
Quick update.
I've been thinking about starting short weekly community calls beginning next week. Just a quick way to keep everyone updated, answer questions, and hear your feedback.
What do you think?
As for the Polyfactual Podcast - we're waiting to finalize a time with their team. Once it's confirmed, I'll post the exact date and time.
No excuses this time. 🙂
- 3 replies
- 9 likes
- 746 views
-
HM, Forecasters. ☀️
Happy Monday. Wishing everyone a productive week.
Today and over the next couple of days, we're focused on launching Proof Network on @robinhoodapp mainnet and finish one more powerful integration.
The Podcast with the @Polyfactual team is still on. As soon as we finalize a time that works for their team, we'll announce it.
Stay tuned.
- 4 replies
- 1 reposts
- 7 likes
- 590 views
-
What we shipped this week. Quick Forecast AI recap:
• Full product redesign
• Clearer outputs across all 7 AI agents
• Improved citations and source provenance
• @tavilyai integration for faster live research and stronger reasoning
• Chat with the Swarm after every analysis
• Live @Polymarket and @Kalshi markets, search and analysis
• Falcon API by @Polygun_ integrated, enriching our 7-agent Swarm with live market intelligence, Smart Money activity and social signals.
• Opportunity Radar and Robinhood RWA signals
• Proof Network live on @RobinhoodApp Chain Testnet
Next up:
• Another intelligence partner integration (already in process)
• Proof resolution, Brier Scores and public agent rankings, followed by the Proof Network launch on @RobinhoodApp Mainnet.
• First steps toward user-created agents
• Expanded $FORAI utility and agentic payments (for token utility putting the final pieces together before sharing the roadmap with the community)
Thanks to everyone who supports us. See you next week.
- 5 replies
- 2 reposts
- 12 likes
- 1.2K views
-
How was your Sunday, Forecasters?
Back to building. Full focus on Forecast AI again.
Lots to ship. I'll keep you updated.- 5 likes
- 346 views
-
HM, Forecasters. ☀️ Happy Sunday.
Wishing everyone a wonderful day with family and friends.
Will be back to work later today.
Stay tuned.
- 2 replies
- 9 likes
- 456 views
-
My apologies to everyone, and to the @Polyfactual team.
Unfortunately, today's Space has to be postponed.
Planning to hold it on Monday. Thank you for your understanding.
I'll make sure it's worth the wait.- 5 replies
- 12 likes
- 667 views
-
Forecast AI is one step closer to becoming a verifiable onchain reputation network.
We are building transparent track records for AI agents analyzing prediction markets, and our Proof Network is now live on the @RobinhoodApp Chain Testnet.
How it works: Before a prediction market resolves, forecasts from all seven agents are timestamped and recorded onchain, so they cannot be changed afterward. Once the official outcome is known, each agent receives a Brier Score and builds a verifiable accuracy ranking.
This infrastructure also lays the foundation for users to create their own forecasting agents using Forecast AI, build verifiable track records, and eventually publish them through an agent marketplace.
This also creates a foundation for expanding $FORAI utility.
Already implemented:
• ForecastRegistry deployed
• First seven-agent forecast recorded
• Automated onchain proof publishing
• Resolution and agent-ranking infrastructure
Example transaction:
https://explorer.testnet.chain.robinhood.com/tx/0x37ab3bd78d2764ae082a3fb17d08c3644dc7f958effb79b6d778629a2a1ffe91
Next: complete resolution testing, public agent track records and Robinhood Chain mainnet deployment.
- 3 replies
- 1 reposts
- 8 likes
- 755 views
-
Happy Saturday, Forecasters.
Touching grass or ...?
We've got one more major update coming today and another integration already underway.
More soon. Enjoy your weekend.
- 2 replies
- 2 reposts
- 11 likes
- 671 views
Forecast AI Reviews & Comments
No reviews yet. Be the first to share your experience.
Forecast AI alternatives
Other Polymarket tools that do a similar job. Compare and pick what fits your workflow.
-
Compare
Ember
Traders who want a second opinion on Polymarket odds get daily AI probability assessments…
-
Compare
Poly Prediction
Traders who want context behind Polymarket prices get AI-driven analysis that connects li…
-
Compare
Polyfactual
Polyfactual combines live prediction market data with AI to give traders and developers i…
-
Compare
Polymarket Institute
Doctoral researchers get funded time, real market data, and open APIs to study how predic…
Recommended reading
In-depth guides for AI Tools — get more out of tools like Forecast AI.
-
Beginners
Polymarket Weather Markets: A Data-First Trading Guide
Learn how to trade Polymarket weather markets with a repeatable workflow: read the station rules, stack ECMWF and METAR forecasts…
8 min read -
Market News
Top 20 Polymarket Influencers And Traders to Follow on X (2026)
A ranked guide to 20 Polymarket influencers on X. Find data analysts, political sharps, and niche traders who share their reasoni…
6 min read -
Tool Reviews
Free Polymarket Open Source Tools and Trading Bots on GitHub
Free Polymarket open source tools and trading bots on GitHub. Datasets, backtesters, copy trading, weather bots, and AI agents, w…
10 min read
Ready to try Forecast AI?
Open the official site and get started in a few clicks.
Leave a comment