Work / Blockchain

Trading-Agent

Autonomous LLM-orchestrated trading on Gryps RFQ perps.

Phase
0 · paused
Joint surface
Trading P&L · dogfooding · research IP
Pause reason
Protocol-side alignment with Gryps hedger

Summary

Trading-Agent is the autonomous LLM-orchestrated trader being built on Gryps RFQ perpetuals. Joint surface area: real trading P&L, dogfooding of the Gryps stack, and research IP for the agentic-systems thread that feeds both Plumbline tooling and the cetacean HITL architecture work.

Why it matters

Most LLM-trading work either runs in toy simulators or wraps existing centralized exchange APIs. The interesting question. Whether an agent can run autonomously on a real on-chain RFQ venue, with real submission paths and real failure modes. Needed an actual on-chain venue built for institutional flow. Gryps is that venue.

What was noteworthy

Phase 0 paused on 2026-05-11 because the agent's open flow was racing the production hedger. A real protocol-side coordination issue, not a tooling problem. Honest framing matters here. The pause is a feature of the joint-development model; the dependency on Gryps protocol decisions is what makes the dogfooding loop work.

Back to Blockchain