Runtime
Each agent is a purpose-built Rust process with a custom Hyperliquid API integration layer. Rust matches Hyperliquid’s own L1 stack and provides both the performance required for real-time scanning and the compile-time safety guarantees critical for a system handling live trade execution.Skill Files
Agents ship with pre-built skill files encoding entry criteria, risk parameters, position sizing, technical analysis methodology, regime detection, and behavioral pattern recognition. A handler receives a working, optimized agent on day one – not a blank canvas requiring prompt engineering. Skill files are versioned and maintained by the core team. As the lesson library grows, validated patterns from the collective feed into skill file updates. When a skill file improves, every agent in the network improves simultaneously.Inference Pipeline
- Bulk data pull – Background polling fetches market-wide data from all 20 intelligence layers into local caches.
- Local filter – Deterministic computation (indicators, funding anomalies, OI shifts, volume spikes) narrows 200+ markets to 5-10 candidates. No LLM involved.
- LLM reasoning – Only the filtered shortlist goes to the language model for deep analysis, proposal generation, and natural language reasoning.
- Handler-directed analysis – The handler can request a deep dive on any specific asset at any time. The agent bypasses the scanner, pulls from cache, fetches fresh data where needed, and delivers a full assessment.