CloddsBot is an open-source AI trading agent that scans prediction markets, crypto exchanges and on-chain DEXs and places trades on its own. It is aimed at developers and technically confident traders who are comfortable running a Node service themselves, holding their own exchange keys and wallets, and letting an automated program move real money without asking first. The project is written in TypeScript, published to npm as clodds, released under the MIT license, and currently sits at 1,962 stars and 273 forks. Its own one-line pitch explains the name: Claude plus odds equals Clodds.
What it does
CloddsBot connects to a wide spread of venues from a single running agent. The repository lists Polymarket, Kalshi, Binance, Hyperliquid, Solana DEXs and five EVM chains, and advertises coverage of more than 1,000 markets, with a dedicated section of the README covering ten prediction markets.
The pitch behind that breadth is simple. A bot pointed at one exchange only sees what that exchange shows it, so mispricings on the other nine venues go unnoticed. CloddsBot's model-driven loop is meant to watch all of them at once, look for edge, execute, and keep managing position risk after the trade is on — including overnight, when nobody is at the keyboard.
Beyond trading it also implements an agent commerce protocol for machine-to-machine payments, tagged in the repository as x402, so the agent can pay other agents for services without a human clearing each transaction. The README's navigation also points to a WebChat interface, a set of channels (the repository is tagged as a Telegram bot), an agent forum and a token launch section; the README carries a Solana contract address for a Clodds token.
How it works
Claude is the decision layer, and the surrounding TypeScript is the plumbing: venue connectors, execution, risk management and the chat front-ends. The README advertises a skill-based architecture with 119 or more skills, which is how capabilities are added rather than by bolting on a new script for each exchange.
Everything runs on your machine or your server. Self-hosting means the agent holds your API credentials and wallet access directly, with no third-party service sitting between the model and the order book. The badges specify Node.js 22 or newer and TypeScript 5.3, so this is a modern-Node deployment, not something to drop onto an old box.
Getting started
Install the clodds package from npm and run it; the project's own framing is that one command gets you from install to a live agent, with an onboarding flow that walks you through setup in the chat interface rather than through a config file you edit blind. From there you talk to it through WebChat or a messaging channel.
Two practical notes before you start:
- You supply the accounts. Because it is self-hosted, the exchange API keys, wallet keys and any funding are yours to create, secure and revoke.
- Read the code first. It is MIT-licensed TypeScript, so nothing stops you from auditing the execution and risk paths before you let it trade — and with a program that spends money unsupervised, that is the whole point of choosing an open-source one.
When to use it / when not
This is a good fit if you want one agent covering prediction markets and crypto venues together instead of stitching several single-venue bots into a cron job, if you want the trading logic in a language you can actually modify, and if self-custody of keys matters more to you than convenience.
It is a poor fit if you want a managed product with support, if you are not prepared to review code that has direct access to your funds, or if you would not run an unattended trading process at all. Two other things belong in the decision. The repository is young — created in January 2026, with its most recent push in September 2026 — so it has not been through many market cycles. And it ships with a launched token and a contract address in the README, plus hackathon branding, which is a set of incentives worth understanding before you treat the project as neutral infrastructure.
Attention it clearly has: the README displays a Trendshift badge and a screenshot claiming 10,746 git clones in fourteen days, and the star count backs that up.
CloddsBot is worth a serious look if you are a developer who already trades across several venues and has been assembling your own tooling to do it. The combination of many connectors, a model-driven decision loop, a readable TypeScript codebase and an MIT license is genuinely hard to find in one repository, and self-hosting means the keys never leave you. Treat it as a framework you audit and adapt rather than a finished product you switch on, be sceptical of the headline market counts until you have verified the connectors you care about, and size your first positions as if you were testing a bot you wrote yourself — because effectively you are.