adrianhihi/helix

Implements a 6-stage PCEC repair pipeline that diagnoses errors, generates fixes via LLM, and stores them in a SQLite Gene Map scored by reinforcement learning—enabling sub-1ms instant fixes on repeated errors without LLM calls. Built on the modular Vial framework, it provides payment-specific adapters for Coinbase, Tempo, and Privy alongside generic HTTP support, with 12 repair strategies (nonce refresh, gas acceleration, transaction splitting) and 7 pre-execution safety constraints. Supports TypeScript, Python, REST API, and CLI tooling including autonomous error discovery via self-play and federated learning across agent instances.