LLM Trader — AI · multi-agent

┌──────────────────┐ │ TRADER · agent │ ┌────────────────┐ │ ▓▓▓▓▓▓▓▓░░ │ ──▶ │ RISK GATEWAY │ ──▶ jupiter │ proposes swap │ │ validate · sign│ └──────────────────┘ └────────────────┘

$ man llm-trader

A committee of LLM trader agents proposes swaps; an orchestrator forwards a deterministic intent to a risk gateway that is the only component allowed to sign or execute. The gateway re-quotes against Jupiter at submit-time, enforces policy, writes a SQLite audit trail, and supports both paper and live modes.

$ grep -i highlight

Committee-of-agents pattern with a hard split between proposers and the signing component
Submit-time quote re-validation prevents stale-quote frontrunning the policy check
Local LLM inference via LM Studio using an OpenAI-compatible endpoint
Operator GUI + Make targets (health, snapshot, cycle, report) for live runbooks