Loading…
Knowledge & Memory
Ground agents in documents, sites, databases and SharePoint with hybrid retrieval, reranking and graph expansion — and remember users across conversations with long-term memory.
Capabilities
BM25 keyword and dense vector retrieval fused with RRF, then reranked for precision.
Entity and relationship expansion over Neo4j for questions that span documents.
Numbered citations with previews. Signed-in users open the cited document at the cited page.
Weak snippets are not cited. The agent says it can't find a grounded answer instead of guessing.
Mem0 recalls user facts across conversations — self-hosted, the memories stay in your infrastructure.
Inspect what was retrieved for any answer and evaluate stores against test questions.
How it works
Pick Qdrant, Azure AI Search or Neo4j, and the embedding model for that store.
Files, URLs, YouTube transcripts, SQL views and SharePoint sites.
Linking a store adds the knowledge_search tool automatically.
Use traces and evals to tune retrieval before you publish.
Enterprise context engine
Query rewriting and intent detection choose between vector, hybrid, reranked, graph, hybrid-graph and structured retrieval — or let AUTO decide. Trust and freshness boosts keep stale content from winning.
Retrieval strategy
Under the hood
The specifics your architects and security reviewers will ask for.
Works with
Every capability shares the same runtime, identity and governance.
Yes. Stores, credentials and audit chains are per organization, and you can self-host the vector store and Mem0 so content and memories stay in your infrastructure.
Retrieval applies a no-grounded-answer floor: if nothing relevant enough is found, the agent says so rather than citing weak snippets.
Book a working session with our team. Bring a real use case — we'll build it with you and show the governance and ROI views alongside.
