search_pages

Hybrid full-text + semantic search across the workspace. Use this for freeform "find anything about X" queries where the agent does not know the exact collection or properties.

For structured lookups within one collection, prefer query_pages.

Arguments

Name	Type	Required	Description
query	string	yes	Search query. Must be at least 2 characters.
collection_id	string	no	Scope the search to a single collection.
limit	number	no	Max results. Positive integer, capped at 50. Default 20.
mode	enum	no	hybrid (default), keyword, or semantic.

Modes

• hybrid — full-text search combined with vector similarity. Best for most queries. The default.
• keyword — pure Postgres full-text search. Best when the query contains identifiers or exact tokens that should match literally.
• semantic — pure vector similarity over embeddings. Best for fuzzy concept matching where the exact words do not appear in the target page.

Returns

A JSON array of ranked search results. The API returns pages ordered by relevance; the exact scoring fields depend on the chosen mode.

Example

{
  "tool": "search_pages",
  "arguments": {
    "query": "onboarding checklist for new hires",
    "mode": "hybrid",
    "limit": 10
  }
}

See also

• query_pages — structured filter/sort within one collection
• read_page — fetch the full body of a hit