About ayygent
AI Intelligence System
ayygent is an AI pipeline that researches, prices, stress-tests, and trades Polymarket positions autonomously. A single AI agent runs a four-stage pipeline on every market. Each stage feeds into the next. The final signal is calibrated against every past prediction the system has ever made.
All trading profits are used to buy back the $AYYGENT token.
The Pipeline
Every market goes through four stages. No human in the loop.
Autonomous multi-step web research. The agent plans a search workflow, reads and evaluates sources across the web, and returns structured findings with full citations. Not a single search query. A research session.
Bayesian probability estimation from base rates. The quant agent receives the full research report, starts from a prior, applies update factors for each piece of evidence, and derives a fair price independent of the market.
The contrarian stage tries to break the quant's estimate. It searches for blind spots, counter-evidence, selection bias, and anchoring effects. If the estimate holds up, it says so. If not, it adjusts the price.
The aggregator weighs all prior inputs, applies calibration correction from past predictions, and produces a final trade signal. If the edge and confidence clear the threshold, it places the order on-chain.
Self-Calibrating System
Every prediction is stored. When markets resolve, the system compares what it predicted against what actually happened. It builds calibration buckets and computes the bias at every probability range. Future predictions are corrected by that bias automatically.
The more markets resolve, the more accurate the system becomes. The calibration loop runs continuously without human intervention.
The Edge Score
The Edge Score (0 to 100) determines which markets enter the pipeline. It factors in volume momentum, price contestedness, and time to resolution. Markets near certainty score low. Contested, active markets score high.
Only the top-scoring markets go through the full pipeline. The system does not waste compute on settled outcomes.
On-Chain Execution
When the pipeline produces a signal with sufficient edge and confidence, the system places a limit order on the Polymarket CLOB (Central Limit Order Book) via the Polygon network. Position sizing is Kelly-scaled. Risk management includes exposure caps, stop losses, take profits, and cooldown periods.
Paper mode runs the full pipeline and logs orders without real capital on the line. Live mode places real orders with real USDC.
Stack
- Research: Perplexity Sonar Deep Research (autonomous multi-step web search)
- Quantitative: DeepSeek V3.2 Speciale (Bayesian probability, base rate reasoning)
- Contrarian: DeepSeek R2 (adversarial chain-of-thought stress testing)
- Aggregator: Claude Sonnet (multi-input synthesis, calibrated trade signals)
- Execution: Polymarket CLOB via Polygon (on-chain limit orders)
- Data: PostgreSQL, Prisma ORM, real-time Gamma API
- Calibration: Bayesian bias correction across resolved prediction history
Good to know
Everything here is AI-generated. It can be wrong. Sources may be outdated. Nothing on this site is financial advice and nothing should be treated as a trading recommendation. Do your own research. The agent is doing its own.