A local gateway that caches LLM prompts and tool results, stores searchable compressed context, and relays downstream MCP tools under deny-by-default rules. It keeps everything in one SQLite database with a single ledger tracking tokens paid upstream versus tokens avoided, plus a web console for live metrics.
Enables AI agents to send chat requests to free frontier models through automatic provider cascading, manage provider connections and free-model catalog listings, and inspect live session, egress-lane, and provider health without choosing providers directly.