The Brain · 04.05
Knowledge Bases
Ingest, chunk, embed, auto-link, and enrich uploaded documents — PDFs, docx, and text — into searchable knowledge atoms.
Scoped like everything else in the Brain
A knowledge base is either shared workspace-wide or scoped to a specific workflow — the same scoping discipline the rest of the Brain follows. Documents are added as raw content or raw bytes (with an optional describeImage step for images, feeding the vision pipeline), then chunked, embedded, auto-linked to related atoms, and — where a Brain enrichment provider is configured — enriched with additional structure before they're searchable.
Three retrieval modes for three kinds of question
| Mode | When to use it |
|---|---|
strict | You want only chunks tightly matching the query — precision over recall. |
contextual | The default balance — relevant chunks plus enough surrounding context to be useful. |
exploratory | You want breadth — casting a wider net across loosely related material. |
await agentis.knowledge.search({ knowledgeBaseId, query: "refund policy", retrievalMode: "contextual" });Continue
An evidence ledger, versioned sources, claims with an approval workflow, conflict resolution, and per-agent grants — the Brain's organizational-intelligence engine.
Body on disk, metabolism in the Brain: a real SKILL.md a harness loads natively, paired with a confidence-tracked skill atom that graduates by reuse.