AI-Trader: HKU's Open Trading Floor for Agents
AI-Trader from HKU's Data Science group hit 15K stars on GitHub, with 200+ added today. Pitch: a fully agent-native trading platform where AI agents — not humans — exchange ideas, share signals, copy each other, and trade. FastAPI backend, React frontend, a separate skills layer that any compliant agent (OpenClaw, nanobot, Claude Code, Codex, Cursor) can read to learn how to participate.
The interesting design call is that the platform is built around agents being first-class citizens. Agents publish signals, debate them in threads, follow each other, and execute trades autonomously. Humans can spectate or copy the bots. Polymarket paper trading went live in March, dashboard in late March, codebase streamlined for agent-native development in April. The team has been shipping monthly.
Why this matters: every prior 'AI for trading' project has been one agent trying to beat the market. AI-Trader is the first serious attempt at a market made of agents, where the price discovery happens between AI participants exchanging information. If the project gets traction, it becomes a standing benchmark for agent reasoning under adversarial conditions, in a way that no static benchmark can match.
Code at github.com/HKUDS/AI-Trader.
← Back to all articles
The interesting design call is that the platform is built around agents being first-class citizens. Agents publish signals, debate them in threads, follow each other, and execute trades autonomously. Humans can spectate or copy the bots. Polymarket paper trading went live in March, dashboard in late March, codebase streamlined for agent-native development in April. The team has been shipping monthly.
Why this matters: every prior 'AI for trading' project has been one agent trying to beat the market. AI-Trader is the first serious attempt at a market made of agents, where the price discovery happens between AI participants exchanging information. If the project gets traction, it becomes a standing benchmark for agent reasoning under adversarial conditions, in a way that no static benchmark can match.
Code at github.com/HKUDS/AI-Trader.
Comments