Prediction markets are notoriously noisy and prone to hype-driven speculation. Polyseer tackles this by orchestrating a multi-agent AI system that digs beneath the surface, systematically gathering and weighing evidence to produce analyst-grade verdicts on Polymarket and Kalshi markets.
how polyseer analyzes prediction markets with specialized AI agents
Polyseer is a multi-agent AI framework focused on analyzing prediction markets, specifically Polymarket and Kalshi. Instead of relying on a single AI model to generate surface-level speculation, it uses a team of specialized agents that collaborate to research, critique, and synthesize evidence.
At its core, the system orchestrates five main agent roles:
- Planner: Designs the research plan and orchestrates the workflow.
- Researcher: Gathers evidence from diverse sources like academic papers, news articles, and market data.
- Critic: Performs gap analysis on the collected evidence, identifying missing information or points needing further inquiry.
- Analyst: Aggregates evidence using Bayesian probability methods, weighting evidence quality with a Type A-D classification system and adjusting for correlations.
- Reporter: Produces the final verdict and report based on the analyst’s synthesis.
The system integrates deeply with the Valyu API for evidence retrieval, enabling robust search across academic and news databases. This helps ensure the evidence base is broad and grounded, not just surface web scraping.
Polyseer is built with modern web technologies: Next.js 16 and React 19 power the frontend and backend, while AI orchestration uses an AI SDK to manage large language models (LLMs). For data persistence in self-hosted setups, it uses SQLite, keeping the footprint lightweight.
Supporting both Polymarket and Kalshi platforms, the project offers a focused solution for prediction market analysis, aiming to reduce bias and improve the quality of market insights.
the multi-agent research loop and bayesian evidence aggregation
What sets Polyseer apart is its structured multi-agent workflow combined with Bayesian aggregation of evidence.
The agents do not just gather confirmatory evidence; they research both PRO and CON viewpoints in parallel. This deliberate balance helps avoid confirmation bias, which is a common failure mode in prediction market commentary.
The Critic agent plays a crucial role by reviewing the initial research results and highlighting gaps or contradictions. This triggers a second research cycle that focuses specifically on those gaps, improving the thoroughness and robustness of the evidence.
Polyseer’s Analyst agent then combines all gathered evidence quantitatively using Bayesian probability aggregation. It weights evidence quality by classifying sources into Type A-D categories (e.g., peer-reviewed papers vs. less reliable news) and adjusts for correlation between evidence items to avoid overweighting related points.
This approach produces probabilistic verdicts that resemble analyst-grade reports rather than simple sentiment or prediction scores.
On the code quality side, the project follows modern TypeScript patterns and maintains a clean separation of concerns among agents. The AI SDK abstracts LLM orchestration, making it easier to swap model providers or update prompts. The use of SQLite for local storage in self-hosted mode is a pragmatic choice, balancing simplicity and persistence.
Tradeoffs include the inherent complexity of managing multiple agents and the reliance on external APIs (OpenAI and Valyu), which requires valid API keys and good network connectivity.
quick start for self-hosted polyseer
The project provides a straightforward self-hosted setup with just three environment variables. Here’s the exact quick start from the README:
git clone https://github.com/yorkeccak/polyseer.git
cd polyseer
npm install
# NEXT_PUBLIC_APP_MODE=self-hosted
### Prerequisites
- Node.js 18+
- npm/pnpm/yarn
- OpenAI API key (for GPT-4o / GPT-5 access)
- Valyu API key (for search capabilities, get at platform.valyu.ai)
This minimal setup lets you run Polyseer locally with your own API keys. The README also details environment variables to configure and API key setup.
Once installed, the system orchestrates the multi-agent research loop automatically, querying prediction markets and presenting evidence-backed verdicts.
verdict: a solid multi-agent AI framework for serious prediction market analysis
Polyseer is worth a look if you’re interested in prediction markets and want an AI system that goes beyond shallow speculation. Its multi-agent architecture with a critic-driven research loop and Bayesian evidence aggregation is a relatively rare approach in open source.
The project does a good job of balancing complexity and usability with modern tech (Next.js, React, AI SDK) and a minimal self-hosted footprint via SQLite.
Limitations include the dependency on OpenAI and Valyu APIs, which means you need valid keys and may face rate limits or costs. Also, the system’s complexity might be overkill if you just want quick sentiment scores.
Overall, Polyseer is a practical framework for developers and researchers aiming to build robust, evidence-based AI analysis of prediction markets. It’s a solid example of applying multi-agent AI and Bayesian reasoning to a real-world problem set.
→ GitHub Repo: yorkeccak/Polyseer ⭐ 648 · TypeScript