Recall

The individual memories that answer a question, best first.

POST /api/v1/recall

Use recall when you want the memories themselves rather than one ready block: to show results in your own interface, pick among them, or build a prompt your own way. For the common case of "give me context for this prompt", use Context.

Request

Field	Type	Required	Description
`query`	string	Yes	The question. Up to 2000 characters
`k`	number	No	Most results to return, 1–20. Default 8

curl https://<your-buddhi-host>/api/v1/recall \
  -H "Authorization: Bearer crtx_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "Which words should we avoid in marketing copy?", "k": 5}'

Response

{
  "results": [
    {
      "id": "5b1e…",
      "title": "Brand voice",
      "kind": "fact",
      "scope": "marketing",
      "score": 0.6123,
      "source": "Notion",
      "url": "https://www.notion.so/…",
      "author": "Priya",
      "timestamp": "2026-04-12T00:00:00.000Z",
      "excerpt": "Summary: Plain and confident. Never use the word \"revolutionary\".\nSource text: Our voice is plain and confident…",
      "summary": "Plain and confident. Never use the word \"revolutionary\"."
    }
  ]
}

Field	Description
`excerpt`	The text a language model should read for this memory: its summary followed by the best-matching part of the original
`summary`	A one- or two-sentence standalone statement of the memory, or `null`

The remaining fields are the same as sources in Context.

What to expect

Results are ordered by how well they answer the question, not just by how similar they look. A memory that only shares words with the question is dropped.
An empty list is a real answer. It means nothing in the key's scopes answers the question.
Fewer than k results is normal. Buddhi returns the memories that answer, not padding. One well-matched result is common.
Superseded memories are never returned. When a fact is updated, only the current version comes back.