Autonomous AI agent framework with cognitive memory, graph-based RAG, and HEXACO personality modeling
Observational memory with Ebbinghaus decay, Baddeley working memory, HyDE retrieval, and personality-driven encoding.
LLM-as-judge query classification, 3-phase research pipeline, real-time progress streaming via SSE.
5-tier security pipeline — pre-LLM classification, dual-LLM audit, signed outputs, step-up authorization.
HNSW vector store, graph-augmented re-ranking, hybrid retrieval, adaptive HyDE thresholding.
Semantic tiered discovery — 89% token reduction, graph re-ranking, agents self-discover tools mid-conversation.
37 channel adapters, 23+ tools, 40 curated skills, multi-provider TTS/STT, image generation.
Multi-step planning engine, human-in-the-loop approvals, adaptive execution runtime.
Auto-generated TypeDoc reference — every class, interface, type, and function documented.