Start building
Choose your path. Both lead to a deployed worker earning ALLO.
Build with AI
You don't need an ML degree. Pick a topic, copy the prompt into your favorite AI coding assistant, and deploy your first worker in under an hour.
Choose a topic and difficulty that matches your comfort level
Claude, ChatGPT, Cursor, Copilot... any AI coding assistant works
Wrap it in Docker, push to Forge, and start earning ALLO
I want to build an Allora worker that predicts ETH/USD price 10 minutes into the future.
Here's what I need:
1. A Python Flask API with a /inference/<token> endpoint
2. Use a simple approach: fetch recent ETH prices from CoinGecko's free API, calculate a weighted moving average of the last 20 candles, and add momentum adjustment based on the last 5 candles
3. Return the prediction as JSON: {"value": "predicted_price"}
4. The server should run on port 8000
Keep it simple. No complex ML libraries needed. Just requests, flask, and numpy.
Include error handling so it returns the last known price if the API call fails."At Numerai's DeAI Day in Vienna, a participant deployed their first Allora worker in under an hour using an AI coding agent."
You don't need to be an ML engineer. You need a good idea and an AI assistant.
Allora Forge: the road to mainnet
Mainnet is whitelist-only, and the whitelist is earned here. You run a worker on a testnet topic inside a Forge competition, the chain scores every submission, and passing the promotion bar below gets that worker onto mainnet. There is no other door.
Nothing is open right now. The Forge runs competitions on an ongoing basis; the next one appears here within an hour of being listed.
Ended: NEAR Inference Grand Prix (100,000 ALLO), BTC 5-Minute Inference Sprint (75,000 ALLO), BTC Price Prediction Sprint (50,000 ALLO).
Create a worker with the Builder Kit, fund it from the testnet faucet, and register it to the topic of an open competition.
Submit an inference every epoch for the whole window. Scoring is on-chain against the topic's ground truth, and each competition pays its own ALLO prize pool.
Promotion is judged per worker, per topic, on confidence intervals rather than point estimates. Pass every metric and the worker is whitelisted for mainnet.
Every metric must pass, per worker and per topic. Where a row says CI, the bound that counts is the confidence interval, not the headline value.
Source: docs.allora.network/build/forge/competitions, read 2026-09-02. Long-horizon topics get a horizon adjustment on the three improvement metrics.
Recommended Data Sources for Your Model
Now let's talk about earning...
How to Earn ALLO →