Overview
Polyseer is an open-source research platform designed to generate systematic, evidence-based analysis for prediction markets. It uses a multi-agent architecture where different agents specialize in different data sources and analysis methods, then aggregates their outputs using Bayesian probability to produce confidence-weighted forecasts.
For agent builders, Polyseer serves as a ready-made Layer 4 intelligence module. Rather than building your own analysis pipeline from scratch, you can use Polyseer’s multi-agent system to evaluate markets and feed the results into your trading logic.
Key Features
Multi-agent analysis — Separate agents handle different data sources (news, social media, historical patterns, expert forecasts). Results are aggregated rather than relying on a single model.
Bayesian probability aggregation — Combines outputs from multiple agents using Bayesian methods to produce probability estimates with mathematical confidence scores.
Cross-platform support — Analyzes markets on both Polymarket and Kalshi, making it useful for cross-platform arbitrage detection.
Report generation — Produces comprehensive analysis reports for individual markets, including probability estimates, confidence intervals, and key factors.
Open source — Fully open source, allowing customization and extension for specific use cases.
How the Multi-Agent Architecture Works
Polyseer doesn’t rely on a single LLM call to evaluate a market. Instead, it runs multiple specialized agents in parallel, each analyzing the question from a different angle:
- News Agent — Scans recent news articles and press releases for relevant information that affects the outcome
- Social Sentiment Agent — Analyzes social media signals from X/Twitter and Reddit for crowd sentiment shifts
- Historical Pattern Agent — Looks at similar past events and their outcomes for base rate estimation
- Expert Forecast Agent — Aggregates predictions from domain experts and forecasting platforms
Each agent produces an independent probability estimate with a confidence score. Polyseer then uses Bayesian aggregation to combine these into a single posterior probability, weighted by each agent’s historical accuracy on similar market types.
Integration Example
from polyseer import MarketAnalyzer
analyzer = MarketAnalyzer(
llm_provider="anthropic",
api_key="your-anthropic-key"
)
# Analyze a specific Polymarket market
report = analyzer.analyze(
market_id="polymarket:condition_id_here",
agents=["news", "social", "historical", "expert"]
)
print(f"Estimated probability: {report.probability:.2%}")
print(f"Confidence: {report.confidence:.2%}")
print(f"Market price: {report.market_price:.2%}")
print(f"Estimated edge: {report.edge:.2%}")
# Use the output to size a bet via Kelly criterion
if report.edge > 0.05 and report.confidence > 0.7:
kelly_fraction = report.kelly_optimal_fraction(bankroll=1000)
print(f"Kelly-optimal bet: ${kelly_fraction:.2f}")
The structured output format makes Polyseer a natural fit for connecting to trading execution layers — pipe the probability estimates and confidence scores into your agent’s betting stack to drive automated position sizing.
Polyseer vs. Building Your Own Intelligence Layer
| Consideration | Polyseer | Custom Build |
|---|---|---|
| Time to production | Hours — install and configure | Weeks — build each analysis pipeline |
| Analysis depth | Good — multi-source, Bayesian aggregation | Customizable — exactly what you need |
| Maintainability | Community-maintained, updates included | You maintain all data source integrations |
| LLM costs | Higher — runs multiple agents per analysis | Lower — optimize to your specific needs |
| Flexibility | Moderate — extend via custom agents | Full — complete control |
For most builders, Polyseer is the right starting point. Build a custom intelligence layer only if you have a specific analytical edge that Polyseer’s architecture can’t capture.
Agent-Friendliness
| Aspect | Rating |
|---|---|
| Output format | Structured reports with probability scores |
| Integration | Can be used as a module in larger agent systems |
| Customization | High — open source, extend with custom agents |
| Data sources | Multiple — news, social, historical, expert |
| License | Open source |
Pricing
Free and open source. Requires LLM API costs for analysis operations (typically $0.05-$0.50 per market analysis depending on the number of agents and model used).
Best For
Agents that need a sophisticated analysis layer without building one from scratch. Particularly strong for agents that need mathematical confidence scores to size their bets (Kelly criterion, etc.).