What is Polyseer?
Polyseer is a Next.js app that reads a live Polymarket or Kalshi market (question, prices, close date and resolution rules), builds a forecasting prompt from it and hands the job to Valyu DeepResearch. You watch the agent's searches and reads as they happen, then get a probability against the market price with a range and confidence level, a suggested side (buy YES, buy NO or no edge), evidence ranked for and against, base rates, catalysts and risks, plus a PDF report and a CSV of factors.
There is no model pipeline of its own: the app is a front end around one external research task. Self-hosted mode needs only a Valyu API key and has no sign-in; a Valyu OAuth mode bills each user's own credits and stores users in Supabase. A hosted version runs at polyseer.xyz.
Is Polyseer worth it?
Polyseer produces well-organised research on a prediction market, and nothing in the repository shows that its probabilities beat the market price. The word "calibrated" comes from the prompt: the research strategy tells the agent to work "like a calibrated analyst" and the output schema asks for "your calibrated probability". We found no backtest, Brier score or record of forecasts against resolved markets in the code or README. A model that is asked to be calibrated is not thereby calibrated.
What it does well is structure. Market facts are pulled from the Polymarket Gamma and CLOB APIs or the Kalshi trade API and written into the query, so the agent anchors on the real resolution rules rather than a paraphrased question. The output is schema-validated JSON, rendered as a verdict card with evidence and rationale, and every forecast link is stable. The project has about 665 stars and 112 forks since August 2025, with the last commit on 14 October 2026.
The catch: the cost table is out of step with Valyu's own prices, and the default deployment is open. The README lists Fast at $0.10, Standard $0.50, Deep $2.50 and Max $15.00 per run; Valyu's docs charge 25, 100, 500 and 2,500 credits, which is about $0.16, $0.63, $3.13 and $15.63 at pay-as-you-go rates (monthly plans bring the per-credit price down). Self-hosted mode has no authentication and no rate limiting in the code, so a public URL lets anyone queue research on your key; Max runs take up to 2 hours each.
Our take: read a Polyseer report the way you would read a sharp analyst's memo, for the evidence list and the resolution-rule details, and hold back on the edge it claims until the tool has a scored record on markets that have already resolved.
Where it earns its place
A part-time Polymarket trader with a handful of open positions
Get the evidence on one market without reading every news story
Paste the market URL on Fast for a report in about 5 minutes at 25 credits, then read the evidence list and resolution rules; treat the probability and side as a prompt for your own judgement.
A forecasting researcher at a university lab
Test whether a deep-research agent adds anything to market prices
Self-host it, run Fast or Standard on a batch of markets before they close, log each probability, and score them with a Brier score after resolution. A hundred Standard runs is 10,000 credits, about $62.50 at pay-as-you-go rates.
Before you commit
- Setup
- Node 20+ and pnpm: clone, `pnpm install`, put `NEXT_PUBLIC_APP_MODE=self-hosted` and `VALYU_API_KEY` in `.env.local`, `pnpm dev`. A first forecast on Fast takes about 5 minutes.
- Integrations
- Polymarket Gamma and CLOB APIs, Kalshi trade API, Valyu DeepResearch with JSON output and deliverables, optional completion email; SQLite locally or Supabase in Valyu mode.
- Data and privacy
- Market data and the research query go to Valyu; history is read back from your Valyu account. The README labels output not financial advice, for entertainment and research purposes only.
- Cost at scale
- Valyu credits per run, by effort: 25 (Fast), 100 (Standard), 500 (Deep), 2,500 (Max), plus 16 credits for a second deliverable. Monthly Valyu plans run $29 to $499.
- Limits
- Deep runs take 30 to 60 minutes and Max up to 2 hours, per the README. Only Polymarket and Kalshi URLs are accepted. Valyu OAuth apps are not yet generally available.
What’s good
- Uses live market data and resolution rules, not a paraphrase
- Evidence for and against, each with a source
- PDF report and CSV of factors per run
- Small codebase; self-hosted mode needs one API key
What’s not
- No published track record on resolved markets
- README cost table is below Valyu's pay-as-you-go prices
- Self-hosted mode has no auth or rate limiting
- Licence stated only in the README; no LICENSE file
Best for
Prediction-market traders and researchers who want a structured, sourced first read on a market and will judge the probability themselves.
Polyseer alternatives
Polyseer: common questions
- Is Polyseer worth it?
- A tidy research report generator for prediction markets that calls its numbers calibrated without ever having been scored. We rate it 3.4 out of 5.
- How much does Polyseer cost?
- Code free (README says MIT). Each forecast is a Valyu DeepResearch task: 25 credits on Fast up to 2,500 on Max, about $0.16 to $15.63 at the pay-as-you-go rate of 1,600 credits per $10.
- What are the downsides of Polyseer?
- No published track record on resolved markets; rEADME cost table is below Valyu's pay-as-you-go prices; self-hosted mode has no auth or rate limiting; licence stated only in the README; no LICENSE file.
- Who is Polyseer best for?
- Prediction-market traders and researchers who want a structured, sourced first read on a market and will judge the probability themselves.
- How much does Polyseer cost for a team or in production?
- Valyu credits per run, by effort: 25 (Fast), 100 (Standard), 500 (Deep), 2,500 (Max), plus 16 credits for a second deliverable. Monthly Valyu plans run $29 to $499.
- Is Polyseer safe for company data?
- Market data and the research query go to Valyu; history is read back from your Valyu account. The README labels output not financial advice, for entertainment and research purposes only.
- What are the best Polyseer alternatives?
- The closest alternatives we have reviewed are TradingAgents and GPT Researcher.
Answers checked Nov 3, 2026.