FAQ
Frequently asked questions
Everything on what polybot is, how agents drive it, and how it stays safe. Still stuck? Ask us directly or read the guides.
Basics
What is polybot? +
polybot is an open-source (MIT) Python framework and CLI for agent-driven trading on prediction markets. It ships first-class domain types for markets, orders, positions and risk, ten pluggable strategies, a native MCP server exposing 25+ typed tools to AI agents, and connectors for Polymarket, Kalshi, Opinion and Binance under one unified risk and position model.
Why prediction markets, and why now? +
Two 2026 curves crossed: prediction markets became one of the most-watched market structures — real volume on election, macro and event contracts across Polymarket, Kalshi and on-chain venues — while LLM agents got good enough to reason over market microstructure, read news and hold a thesis. The obvious next step is letting an agent trade these markets; the hard part is doing it safely. That is what polybot is built for.
Who is polybot for? +
Quants and traders who want a multi-venue framework without building exchange plumbing; agent developers who want an MCP-native target for Claude, GPT and custom agents; and engineering teams who need agent-driven trading to be auditable, type-safe and self-hostable.
Is polybot free and open source? +
Yes. polybot is MIT-licensed and published on GitHub and PyPI. You self-host it — your infrastructure, your keys, your data.
What are prediction markets? +
Prediction markets let you trade contracts that pay out based on the outcome of a real-world event (an election, a rate decision, a sports result). Prices behave like probabilities. polybot trades them programmatically across Polymarket, Kalshi and Opinion.
Why not just build a bespoke bot per venue? +
You can — and for a single venue with no agent, that may be simpler. But a bespoke script re-derives markets, orders, positions and risk each time, bolts risk checks onto each bot, and needs custom prompt glue for any agent. polybot gives you one typed framework across four venues, shared risk enforced once, and a native MCP server — so you are not maintaining N drifting scripts.
Agents & MCP
Why does polybot ship an MCP server? +
MCP (Model Context Protocol) lets any compliant AI client — Claude Desktop, custom agents, IDE extensions — call typed tools on polybot. You get market analysis, strategy assessment and approval-gated execution as an agent capability rather than a prompt-engineering project.
What can an agent actually do through polybot? +
Read-only tools let it list and inspect markets, positions, orders and strategy code. Assessment tools let it analyse and compare strategies over a window and suggest improvements. Execution tools are approval-gated: an agent can propose a config change or an order, but a human approves it at the CLI before anything goes live.
Can the agent trade without my approval? +
No. Execution runs in one of three modes — disabled, shadow (paper) and live (approval). Even in live mode, agent-initiated trades require explicit approval per order or against pre-authorised thresholds, and every action is audit-logged.
Strategies & venues
How many strategies are built in? +
Ten: arbitrage, statistical arbitrage, AI model, spread farming, copy trading, resolution arbitrage, calendar spread, momentum, poll divergence and volume spike. Each is a subclass of BaseStrategy, so you can read the source, fork it, or write your own.
Which venues does polybot support? +
Polymarket (on-chain CLOB, USDC collateral, EIP-712 orders), Kalshi (CFTC-regulated US exchange), Opinion (DEX-style long-tail markets) and Binance (spot/perps for hedging). They sit behind a BaseVenue abstraction, so adding a venue does not touch strategies.
Can I plug in my own AI model? +
Yes. AIModelPlugin is a base class: return a probability and a confidence and polybot handles signal generation, sizing and risk gating. Claude, GPT and Perplexity ship built-in; point at any HTTP endpoint or local model for your own.
Safety & operations
Is polybot production-ready? +
It is beta (0.1.x), MIT-licensed, strictly type-checked with mypy, tested, and deployed via Docker. Risk controls — position caps, daily loss limits and shadow-first defaults — are built in rather than bolted on.
How does shadow mode work? +
Every strategy boots in shadow (paper) mode: it generates signals and simulates fills without touching real funds. You promote a strategy to live explicitly and per-strategy once its metrics prove out, and you can roll it back just as easily.
What risk controls are enforced? +
Per-market, per-category, per-venue and global USD exposure caps, plus daily loss limits that auto-disable a strategy or venue on drawdown. These are enforced by the platform before an order is submitted — not by each strategy.
Will polybot make me money? +
polybot makes no return claims — it is infrastructure for trading prediction markets carefully, not a strategy for profit. Prediction-market trading carries real financial risk, and every strategy can lose money. The safety controls (paper-trading default, human approval, risk caps) exist to help you manage that risk, not to promise an outcome.
Where does polybot store data? +
A SQLite operational store holds markets, orders, positions, trades and configs with ACID transactions; a DuckDB analytics store holds price history, trade history and strategy stats for millisecond OLAP queries. Prometheus metrics scrape into Grafana.
Still have questions?
Ask the team, or read the docs. We answer real deployment questions — email contact@cryptuon.com.