Search Brain records by a query, optional built-in/custom item-type filters, and a bounded result count, then emit both structured chunks and formatted context.
Where: Flowgen → Add → Words & data → Knowledge Search
Key ideas
- Retrieval versus full corpus: Knowledge Search retrieves the most relevant chunks for one query. Context materializes a whole connected corpus, and Brain Read binds one specific record.
- Type-aware scope: The card/sidebar can filter by merged built-in and custom Brain item-type ids; no selection means all types.
- Two outputs: chunks_out contains structured retrieved hits; context_out contains a formatted markdown context block for text/AI consumers.
- Card and sidebar parity: Query, type chips, and result limit (4, 8, 12, or 20) are available in the node card/dedicated research-family inspector, with structured preview for results.
- Knowledge Search card: KnowledgeNode uses the generic shared-card path. Retrieve relevant chunks from Brain records and emit both structured hits and formatted context.
- Properties sidebar: researchNodes.tsx
- Opened surfaces: Structured result preview
Steps
- Enter or connect the query.
- Choose the Brain item types that may answer it and set a small result limit.
- Run and inspect source/title metadata and excerpts in the structured results.
- Connect context_out to a prompt/agent or chunks_out to structured processing.
Knowledge Search controls and presets
- Add path: Flowgen → Add → Research & Insights → Knowledge Search.
- Card: KnowledgeNode.
- Properties: researchNodes.tsx.
- Controls: Query; built-in/custom type filters; limit 4/8/12/20; chunks; context.
- Opened surfaces: Structured result preview.
Limitations and important notes
- Retrieval quality depends on stored content, item types, query wording, and indexing; inspect sources before relying on the context.