Context Engine
Structured knowledge retrieval for AI coding agents. Query codebase context in milliseconds — no external services, no network dependencies.
219 Context Docs
8,251 Triggers
1,711 Sections
27 Topic Clusters
Structured Knowledge Layer
- Context documents describe every system — rendering, physics, networking, UI, simulation — with trigger phrases tuned for search
- Each document is divided into sections with dedicated trigger phrases, topic clusters, and graph edges to related documents
- Triggers are multi-word phrases designed to match the kinds of queries AI agents actually ask — API names, task descriptions, synonym expansions
Two-Step Discovery
- Browse returns ranked section titles with relevance scores — a discovery step to find what exists
- Fetch retrieves full section content plus one hop of graph expansion, pulling in related context automatically
- Hybrid search combines keyword matching with semantic similarity — finds relevant context even when query vocabulary doesn't overlap with trigger phrases
Self-Improving Knowledge
- Completed orchestrator tasks generate new context documents, feeding knowledge back into the system
- Knowledge gaps detected during planning trigger research agents that enrich the context base before execution begins
- Every task makes the next task smarter — the agent that builds features also builds the knowledge to build them better