omnia-odds: prediction-market odds for agents

Balance: address=<0x..> on Base. ETH and USDC balances with raw units and block height.

Chat: q=<prompt> (optional system=<instruction>) to a fast open LLM (Llama 3.1 8B); plain-text answer up to 400 tokens.

AI-judged YES consensus across Polymarket, Manifold and Kalshi for q=<question>: an LLM judge keeps only markets that resolve on the same event (look-alikes and buckets are listed under excluded), then a volume-weighted mean with min/max/disagreement.

Extract: url=<public page>, q=<question or "fields: a, b, c">. Fetches and reads the page, answers with verbatim quotes; found=false when absent.

Quick forecast for q=<yes/no question>: probability, rationale, drivers and what to watch, anchored on live market prices where they exist.

Web read: url=<http(s) page>. Returns clean text (up to 120 KB), title, meta description, links; JSON and plain text pass through.

Cross-venue prediction-market search: q=<text>, limit<=50. Returns venue, question, YES price, outcomes, volume, traders, close date, url.

Token risk report: address=<Base ERC-20 contract>. Honeypot simulation, taxes, liquidity, volume, holders, verification, pair age; grade BASED/MID/COOKED/RADIOACTIVE with risk_score 0-100, reasons and a model-written take.

OpenAI-compatible chat completion: JSON body {model: "fast"|"smart", messages[], max_tokens, temperature}. fast = Llama 3.1 8B, smart = Llama 3.3 70B. Standard chat.completion response.