What is AgentNash?

AgentNash is an autonomous prediction market trading platform. Deploy AI agents that scan live markets, run structured analysis, and execute trades on real exchanges — 24/7, with no manual intervention. Every decision flows through a deterministic safety pipeline before capital is ever committed.

Agents trade on Kalshi (a CFTC-regulated US exchange) and Polymarket (a decentralized prediction market on the Polygon blockchain). Each exchange has its own credentials, market structure, and settlement rules, but the AI analysis pipeline is identical across both — deploy the same strategy on either exchange with a single click.

The flagship strategy is Council V2, a sequential 5-agent adversarial debate. Perplexity gathers live web research, then five specialized agents — Forecaster (Grok 4.1 Fast), Bull Researcher (Claude Opus 4.6), Bear Researcher (Claude Sonnet 4.6), Risk Manager (Claude Opus 4.6), and Trader (Claude Sonnet 4.6) — each build on the previous agent's output before the Trader makes the final call. A trade only executes when the Risk Manager approves and the Trader confirms edge. Also available: Superforecaster, a research-first strategy where Perplexity gathers comprehensive evidence and a single reasoning model applies structured decomposition with base rates to produce calibrated probabilities.

Core Concepts

• Multi-Model Debate — Five AI agents from two providers (xAI and Anthropic) argue through a sequential pipeline where each agent sees and challenges the previous output. No groupthink — disagreement is built in.
• Multi-Exchange Support — Trade on Kalshi and Polymarket through dedicated bot services. Same AI pipeline, different execution layers.
• Training Mode — The full pipeline runs end-to-end but no real orders are placed. Test strategy performance risk-free before committing capital.
• Live Trading — Real orders placed on exchanges through their APIs using your own account credentials.
• Transparency — Every decision is logged with full reasoning from each agent, weighted confidence scores, and rule validation results. Nothing is a black box.
• Two-Tier Safety — Per-agent rules and account-level rules validate every trade. Nine hard constraints are checked — the more restrictive limit always wins.
• Emergency Controls — Stop all agents instantly, or trigger a nuclear shutdown that halts trading and wipes all stored credentials.
• Agent Management — Monitor running agents in the Terminal, review deployed strategies on the Agents page, and compare historical performance on the Benchmarking leaderboard.
• Guided Walkthrough — A built-in interactive tour walks you through every section of the platform — strategies, trades, portfolio, and settings — so you can get oriented before deploying your first agent.