An AI trading agent built in 12 days with an everything-bagel problem

What it does Clodds is a self-hosted AI trading terminal that wraps Claude around a sprawling mesh of betting and financial infrastructure. You chat with it in a browser or over messaging apps, and it routes trades across prediction markets like Polymarket and Kalshi, centralized futures exchanges such as Binance and Hyperliquid, and on-chain venues on Solana and five EVM chains. The README says it was built in twelve days for a Solana hackathon, which accounts for the kitchen-sink feature list: whale tracking, arbitrage detection, DCA bots, token launches, Bittensor mining, and even an agent-to-agent marketplace.

The interesting bit The project treats natural conversation as the primary interface for high-leverage finance. Instead of a Bloomberg terminal, you get a Claude-style sidebar with context compacting and unlimited history, while the backend juggles risk engines, MEV protection, and cross-chain USDC payments via the x402 protocol. It is essentially an operating system for an autonomous trading agent, complete with an MCP server to expose its 119 skills to Claude Desktop.

Key highlights

• Supports 10 prediction markets, 7 perpetual-futures exchanges, Solana DEXs, and 5 EVM chains from a single chat interface.
• Bundles 118+ trading strategies, a unified risk engine with circuit breakers, and backtesting infrastructure.
• Includes a built-in WebChat with semantic memory, context compacting, and persistent SQLite logging.
• Exposes all skills as an MCP server and supports machine-to-machine USDC payments via the x402 protocol.
• Claims 10.7k Git clones in fourteen days and a twelve-day build window for the Colosseum Agent Hackathon.

Caveats

• The README lists an exceptionally broad scope—119 skills, 21 messaging platforms, and 16+ trading venues—alongside a claim that the project was built in twelve days, which raises questions about current depth.
• Several advanced features, such as Percolator on-chain perpetuals and Bittensor subnet mining, are documented primarily via configuration flags and command examples rather than architectural detail.
• There is no disclosed track record, third-party audit, or live performance data for its trading strategies or risk engine.

Verdict Developers curious about autonomous agents or prediction-market automation will find a dizzying reference architecture here. Anyone looking for a battle-tested, audited trading system should wait for harder evidence than a feature list.