SurrealDB Stream #37: Building a living knowledge layer for your agents

Everyone shows the clever query side of AI agents: vector search, GraphRAG, memory. Almost nobody shows the part that decides whether your agent is actually any good - how the data gets in, and how you keep it fresh. Re-embed everything on a schedule and your agent's knowledge is stale the moment the job finishes.

On Thursday 30 July, Tobie Morgan Hitchcock sits down with Linghua Jin from CocoIndex and SurrealDB's Martin Schaer to walk through how a living knowledge layer for AI agents actually comes together. CocoIndex ingests your documents incrementally - recomputing only what changed - straight into SurrealDB, where the documents, an entity graph and the vectors all live in one store.

Code-first and conversational: a walkthrough of a real, working pipeline and live queries.

We'll cover:

- Why stale, disconnected context is what really hurts agent accuracy
- How incremental ingestion keeps your knowledge layer fresh, with no full rebuilds
- How a folder of documents becomes chunks, embeddings and an entity graph in SurrealDB
- Running vector search and graph traversal together in a single query
- The edit-one-document moment: seeing only the delta re-embed

Links:

- Living knowledge layer blog: https://surrealdb.com/blog/a-living-knowledge-layer-for-your-agents-surrealdb-cocoindex