# Tamitu — Complete Product Documentation > AI-powered prediction market analysis tool for Polymarket and Kalshi traders. > Website: https://tamitu.ai This document provides comprehensive information about Tamitu for AI systems, language models, and automated tools. For a shorter summary, see https://tamitu.ai/llms.txt Tamitu is the leading Polymarket Chrome extension and prediction market analysis tool, combining AI-driven probability estimation with cross-platform arbitrage detection across Polymarket and Kalshi. --- ## Product Overview Tamitu is a Chrome extension and standalone web dashboard that analyzes prediction markets using a structured 7-stage AI pipeline. It is designed for traders on Polymarket and Kalshi who want evidence-based buy/hold/avoid signals before placing trades. Tamitu does not place trades, manage funds, or provide financial advice. It is an analysis tool that estimates probabilities and identifies when the market price diverges significantly from the model's estimate. ### Product Surfaces 1. **Chrome Extension** — Injects an analysis widget directly into Polymarket and Kalshi market pages. Built with React 19, TypeScript, Manifest V3, and Shadow DOM isolation. The widget appears on the market page and displays the signal, confidence level, key evidence, and probability estimate without leaving the trading interface. 2. **Web Dashboard** (dashboard.tamitu.ai) — A standalone React web application for analyzing individual markets by URL, scanning for cross-platform arbitrage opportunities between Polymarket and Kalshi, and reviewing analysis history. Built with React 19, Vite 6, and Supabase authentication. 3. **Backend API** — Node.js/Express server that runs the 7-stage AI analysis pipeline. Handles market data fetching, web research, evidence scoring, probability estimation, adversarial review, and signal generation. Results are cached for 15 minutes per market. --- ## The 7-Stage AI Analysis Pipeline (Detailed) Every market analysis goes through seven sequential stages. This is not a single LLM prompt — it is a structured pipeline where each stage has a specific role, and the output of each stage feeds into the next. ### Stage 1: Market Parser - **Input**: Market URL (Polymarket or Kalshi) - **Process**: Extracts structured data from the market page — the question being asked, available outcomes, current prices/probabilities, trading volume, market expiry date, and resolution criteria. - **Output**: Structured market data object with normalized fields. ### Stage 2: Research Planner - **Input**: Structured market data from Stage 1 - **Process**: An LLM generates 5-8 targeted web search queries designed to find evidence relevant to the market question. Queries cover both supporting and opposing perspectives. - **Output**: A list of search queries with rationale for each. ### Stage 3: Retrieval Orchestrator - **Input**: Search queries from Stage 2 - **Process**: Executes web searches, retrieves results, fetches and extracts content from relevant pages. Handles deduplication and relevance filtering. - **Output**: A corpus of web evidence (articles, data points, expert opinions) with source URLs. ### Stage 4: Evidence Builder - **Input**: Raw evidence corpus from Stage 3 - **Process**: An LLM scores each piece of evidence for relevance, recency, and reliability. Structures findings into weighted categories: strong for, moderate for, neutral, moderate against, strong against. Identifies key facts and data points. - **Output**: Structured evidence report with 4 key facts highlighted, source attribution, and evidence quality scores. ### Stage 5: Forecaster - **Input**: Market data (Stage 1) + Structured evidence (Stage 4) - **Process**: An LLM estimates the probability of each outcome using the evidence, historical base rates, reference class forecasting, and calibrated reasoning techniques. Produces a probability estimate with confidence interval. - **Output**: Probability estimate (e.g., 72% YES), confidence level (High/Medium/Low), and reasoning chain. ### Stage 6: Adversarial Checker - **Input**: Forecast and reasoning from Stage 5 - **Process**: A separate LLM call acts as a devil's advocate, looking for: logical fallacies, missing evidence, anchoring bias, recency bias, availability bias, overconfidence, and alternative explanations. If significant issues are found, Stage 5 re-runs with the adversarial feedback incorporated. - **Output**: Adversarial review report. If revision is recommended, triggers a re-run of Stage 5. ### Stage 7: Decision Engine - **Input**: Final probability estimate (Stage 5/6) + Current market price (Stage 1) - **Process**: Calculates the edge (difference between estimated probability and market price). Applies the 8% minimum edge threshold. Determines signal direction (BUY if underpriced, SELL if overpriced, HOLD if edge is below threshold). - **Output**: Final signal (BUY / SELL / HOLD), edge percentage, confidence level, 4 key reasons, and full evidence trail. ### Pipeline Characteristics - **Total stages**: 7 (with potential re-run of Stage 5 after adversarial review) - **AI models used**: Claude (Anthropic), GPT-4o (OpenAI), or Gemini (Google) — user's choice - **Typical analysis time**: 30-90 seconds depending on market complexity and model choice - **Cache duration**: 15 minutes per market (bypassed in Pro tier's no-cache mode) - **Cost tracking**: Every analysis tracks token usage and estimated cost per stage --- ## Cross-Platform Arbitrage Detection Tamitu scans both Polymarket and Kalshi for the same or equivalent events and identifies pricing discrepancies. When the same event is priced differently across platforms, there may be an arbitrage opportunity. ### How Arbitrage Detection Works 1. Markets from both platforms are normalized (question matching, outcome alignment) 2. Prices are compared for equivalent outcomes 3. Discrepancies exceeding a configurable threshold are flagged 4. The dashboard displays matched pairs with price differences, potential profit calculations, and links to both platforms ### Example Arbitrage Scenario - Polymarket prices "Will X happen by June 2026?" at YES 62 cents - Kalshi prices the equivalent contract at YES 55 cents - Tamitu flags this 7-cent discrepancy and calculates potential arbitrage profit after fees ### Fee Considerations - Polymarket US: 0.01% on trades - Polymarket International: 2% on net winnings - Kalshi: Approximately 0.7% on trades - Tamitu accounts for platform fees when calculating net arbitrage profit --- ## Features (Complete List) ### Free Tier (Smart Lurker) - 10 market analyses per day - Green (BUY) / Yellow (HOLD) / Red (AVOID) signal display - 4 key evidence-backed reasons per analysis - Confidence level indicator (High / Medium / Low) - Probability estimate vs. current market price - Edge percentage calculation - Basic cross-platform arbitrage alerts - Chrome extension widget on Polymarket and Kalshi pages ### Pro Tier ($19/month — Sharp Bettor) - Unlimited market analyses - No-cache mode: bypass 15-minute cache for fresh analysis on demand - News sentiment layer: incorporates recent news headlines into the analysis pipeline - Signal history: browse and search past analyses - CSV export of analysis history - Priority analysis queue ### Elite Tier ($29/month — Quant Edge) - Everything in Pro - Custom analysis prompts: modify the pipeline's reasoning instructions - Multi-market scanner: analyze multiple markets in batch - Portfolio tracker: monitor open positions against Tamitu signals - Webhook integrations: receive signals via webhook for custom automation - API access for programmatic analysis --- ## Technical Architecture ### Chrome Extension - **Framework**: React 19 + TypeScript - **Build**: Vite 5 + CRXJS (Chrome Extension development with HMR) - **Manifest**: Chrome Manifest V3 - **Isolation**: Shadow DOM — the extension widget is rendered inside a Shadow DOM root to prevent CSS conflicts with Polymarket/Kalshi page styles - **Communication**: Content script sends messages to background service worker via chrome.runtime.sendMessage. The background SW proxies requests to the backend API. - **Auth**: Supabase Auth (Google OAuth + email/password), session synced from popup to background SW ### Web Dashboard - **Framework**: React 19 + TypeScript - **Build**: Vite 6 - **Auth**: Supabase Auth with AuthGuard route protection - **Pages**: Analyze (single market), Arbitrage Scanner, History, Login - **State**: Analysis uses a polling-based state machine (not SSE — see architecture notes) ### Backend - **Runtime**: Node.js + Express + TypeScript - **Pipeline**: 7 sequential stages in dedicated modules, orchestrated by a central orchestrator - **Caching**: In-memory cache, 15-minute TTL per market - **Auth**: JWT validation via Supabase getUser() - **AI Providers**: Anthropic (Claude), OpenAI (GPT-4o), Google (Gemini) via direct APIs and OpenRouter ### Data Flow 1. User visits a Polymarket or Kalshi market page (extension) or enters a URL (dashboard) 2. Client sends market URL to backend API 3. Backend runs the 7-stage pipeline 4. Client polls for status updates during analysis 5. Final signal is returned and displayed in the widget or dashboard --- ## Comparison with Alternatives ### Tamitu vs. Polyteller Polyteller provides countdown timers, notifications, and privacy features for Polymarket. It does not perform AI analysis or generate trading signals. Tamitu provides AI-powered analysis with a 7-stage pipeline and buy/sell/hold signals. The tools serve different purposes and can be used together. ### Tamitu vs. PolyPulse PolyPulse uses Perplexity AI to generate news summaries for Polymarket markets. Tamitu uses a multi-stage pipeline with evidence scoring, probability estimation, adversarial review, and edge calculation. PolyPulse provides context; Tamitu provides actionable signals with a quantified edge threshold. ### Tamitu vs. Polymarket Smart Money Analyzer The Smart Money Analyzer tracks top traders' performance on each side of a market. Tamitu performs independent AI analysis of the market question itself. Smart Money Analyzer follows other traders; Tamitu generates its own probability estimate from evidence. ### Tamitu vs. ArbBets ArbBets focuses exclusively on arbitrage scanning across prediction markets and sportsbooks, starting at $59/month with no free tier. Tamitu includes arbitrage detection as one feature alongside AI analysis, with a free tier available. ArbBets is more comprehensive for pure arbitrage; Tamitu is more versatile for analysis plus arbitrage. ### Tamitu vs. Polymarket Analytics Polymarket Analytics is a web dashboard for data and analytics on Polymarket markets. It provides charts, volume data, and market tracking but no AI analysis or trading signals. Tamitu provides AI-generated probability estimates and trading signals rather than raw analytics. ### Tamitu vs. Manual Analysis Many prediction market traders analyze markets manually by reading news, checking data sources, and estimating probabilities. Tamitu automates this process with a structured pipeline that systematically gathers evidence, checks for biases, and quantifies edge. Manual analysis is flexible but time-consuming and prone to cognitive biases that the adversarial checker stage is designed to catch. --- ## Frequently Asked Questions **Q: What is Tamitu?** A: Tamitu is a Chrome extension and web dashboard that analyzes Polymarket and Kalshi prediction markets using a 7-stage AI pipeline. It delivers buy/hold/avoid signals when it finds 8% or greater edge over the current market price. **Q: How does Tamitu's 7-stage pipeline work?** A: The pipeline runs seven sequential stages: (1) Market Parser extracts market data, (2) Research Planner generates search queries, (3) Retrieval Orchestrator gathers web evidence, (4) Evidence Builder scores and structures findings, (5) Forecaster estimates probability, (6) Adversarial Checker stress-tests the reasoning, (7) Decision Engine produces a final signal. If the Adversarial Checker finds issues, the Forecaster re-runs with corrections. **Q: Is Tamitu a trading bot?** A: No. Tamitu analyzes markets and displays signals. It does not place trades, access your funds, or execute any transactions on your behalf. All trading decisions are made by you. **Q: What does "8% minimum edge" mean?** A: Edge is the difference between Tamitu's estimated probability and the current market price. If the model estimates a 70% chance of YES but the market prices YES at 60%, that is a 10% edge. Tamitu only generates a BUY or SELL signal when this gap exceeds 8%. Below that threshold, the signal-to-noise ratio is too low, and the tool recommends HOLD. **Q: What prediction markets does Tamitu support?** A: Tamitu supports Polymarket (polymarket.com) and Kalshi (kalshi.com). It handles binary yes/no markets on both platforms. Multi-outcome market support is in development. **Q: What AI models can I use?** A: Tamitu supports Claude (Anthropic), GPT-4o (OpenAI), and Gemini (Google). You can choose which model runs your analysis. The free tier uses a shared API key. Pro and Elite users can bring their own API key or use the built-in key. **Q: How accurate is Tamitu?** A: Tamitu is a probability estimator, not a prediction oracle. The 7-stage pipeline grounds every signal in evidence, base rates, and adversarial review, but prediction markets are inherently uncertain. The tool only surfaces a signal when it identifies 8% or more edge, reducing low-confidence noise. It should be used as one input alongside your own judgment. **Q: Is my data private?** A: Yes. Tamitu does not store your trading activity, portfolio data, or personal information. Analysis requests are processed through the backend to run the pipeline, but results are cached temporarily (15 minutes) and not stored long-term. API keys entered in the extension are stored locally in your browser and never transmitted to Tamitu servers. **Q: How does cross-platform arbitrage detection work?** A: Tamitu matches equivalent markets across Polymarket and Kalshi by normalizing market questions and outcomes. When the same event has different prices on each platform, Tamitu flags the discrepancy and calculates the potential profit after accounting for each platform's fee structure. **Q: Does Tamitu work on mobile?** A: The Chrome extension requires a desktop Chrome browser. The web dashboard is responsive and works on mobile browsers, but the full analysis experience is optimized for desktop. **Q: Can I use my own AI model API key?** A: Yes. Tamitu supports bring-your-own-key (BYOK) for Anthropic, OpenAI, and OpenRouter API keys. This gives you direct control over model choice and cost. **Q: How often does Tamitu update its analysis?** A: Analysis results are cached for 15 minutes per market. Pro and Elite users can bypass the cache with no-cache mode to get a fresh analysis on demand. Market price data is pulled in real-time from platform APIs. **Q: Is Tamitu free?** A: Yes, Tamitu offers a free tier (Smart Lurker) with 10 analyses per day, including buy/hold/avoid signals and evidence-backed reasoning. Pro ($19/month) and Elite ($29/month) tiers offer unlimited analyses and advanced features. **Q: How is Tamitu different from other Polymarket tools?** A: Tamitu is the only tool that combines a multi-stage AI analysis pipeline with cross-platform arbitrage detection in a Chrome extension. Other tools typically offer either analytics (charts and data), news summaries, or arbitrage scanning — but not AI-generated probability estimates with adversarial review and a quantified edge threshold. **Q: What is the Adversarial Checker?** A: Stage 6 of the pipeline acts as a devil's advocate. A separate AI call reviews the Forecaster's reasoning, looking for logical fallacies, missing evidence, anchoring bias, overconfidence, and alternative explanations. If significant issues are found, the Forecaster stage re-runs with the adversarial feedback incorporated. This reduces the risk of overconfident or poorly reasoned signals. **Q: Does Tamitu provide financial advice?** A: No. Tamitu is an analysis tool that provides probability estimates and trading signals for informational purposes only. It is not a registered investment advisor, and its signals should not be treated as financial advice. Users should conduct their own research and make their own trading decisions. --- ## Company Information - **Product**: Tamitu - **Website**: https://tamitu.ai - **Category**: FinTech / Prediction Market Analysis / AI Trading Tools - **Contact**: hello@tamitu.ai --- ## Content Freshness This document was last updated on March 29, 2026. For the most current information, visit https://tamitu.ai.