RAG Memory Configuration
Memory benchmarks (full N=500, gpt-4o reader) round Top-5 65.7%, session Top-5 71.4%, round Top-10 72.0%). Benchmarks · Run JSONs · SOTA writeup
HyDE Retrieval
HyDE improves RAG and memory retrieval by generating a hypothetical answer before
Citation Verification
Per-claim citation verification for AI agents: decompose answers into atomic claims, embed against retrieved sources, verdict ladder (supported/weak/unverifiable/contradicted) with optional NLI contradiction check and web-search fallback.
Multimodal RAG (Image + Audio)
Memory benchmarks (full N=500, gpt-4o reader): 85.6% on LongMemEval-S at $0.0090 per correct, +1.4 points above Mastra Observational Memory (84.23%). 70.2% on LongMemEval-M on the 1.5M-token / 500-session haystack variant — the only open-source library on the public record above 65% on M with publicly reproducible methodology. The same text-first retrieval pipeline that produced these numbers is what the multimodal pattern below indexes against (derived captions, transcripts, OCR, document text) once you have a text representation. Benchmarks · Run JSONs · SOTA writeup
Document Ingestion
Ingest PDFs, DOCX, HTML, Markdown, CSV, JSON, YAML, text, and URLs into the agent memory system with configurable chunking and multimodal image extraction.
Incremental Vector Ingestion
Keep a raw IVectorStore collection in sync with content-hash caching: skip unchanged chunks, re-embed only what changed, and re-run the same ingest as often as you like for near-zero cost. The lower-level path beneath Memory.ingest().