Chunks
Not all work is structured. Tickets and pull requests map cleanly onto entities, but documents, threads, and meeting transcripts are long and free-form. Virgo splits that content into chunks: smaller, retrievable passages that can be searched and returned on their own.
Why chunks exist
When a retriever answers a question, you want the specific passage that matters, not a whole fifty-page document. Chunks make retrieval precise. Each chunk is small enough to return as evidence and is linked back to the entities it mentions, so a passage stays connected to the decision, person, or system it is about.
How chunks are produced
Chunking happens during ingestion, after content is fetched and before it is linked into the graph. The default behavior works without configuration, and you can change it when a source needs different handling. See Chunking hooks to customize how content is split, and Entity extraction hooks to control what is pulled out of each chunk.
Related
- Entities: the nodes chunks attach to.
- Ingesting data: how content reaches Virgo in the first place.