Summary
This video puts Kimi’s open-weights model to a practical test by giving the Kimi Code agent a $10,000 trading account and 22 hours to turn a profit. The creator walks through five steps: installing Kimi Code on desktop, connecting the tools it needs, running the live trading challenge, building an AI trading desk, and reviewing the results.
The setup connects the agent to a broker through a GitHub skill and the trigger.trade relay, which passes the agent’s signals to the broker using API keys. The creator also notes that Kimi Code resembles Claude Code and OpenAI’s Codex but is open source and open weight, and mentions subscription pricing that ranges from $15 to $159 per month for those who cannot run the model locally.
In the second challenge, the agent connects to tradingkit.com through an MCP server to backtest and generate strategies in Pine Script, which can then be verified in TradingView or run live. A test backtest on BTCUSDT confirms the connection, returning a 27% win rate, a 0.9 profit factor, and a 61% maximum drawdown.
The video is useful for anyone curious how coding agents can be extended with skills and MCP servers to operate in real-world, high-stakes workflows without writing code.
📺 Source: Trading with DaviddTech · Published October 29, 2026
🏷️ Format: Workflow Case Study
