prompt_context

Assemble a token-budgeted context block from memories most relevant to a query. Greedy-packs ranked memories within the token budget (1 token ≈ 4 chars) and returns the formatted block plus token accounting metadata. Use it when filling a fixed token allowance in a prompt template; compress_context is the character-budgeted variant. Every response carries retrievalMode: normally “semantic”, but “lexical” when no embedding provider is configured, in which case memories are matched by keyword, minScore is scored on that different scale, and a thin or empty result is not evidence that nothing relevant is stored.

Auth mode: identity
Required scope: memories:read

Input parameters

Parameter	Type	Required	Default	Description
userId	string	yes	—	—
query	string	yes	—	—
tokenBudget	integer	no	2000	—
limit	integer	no	20	—
minScore	number	no	0.5	—
scope	string	no	—	—
tags	string[]	no	—	—
createdFrom	object	no	—	—
createdTo	object	no	—	—

Example

{
  "name": "prompt_context",
  "arguments": {
    "userId": "qp",
    "query": "<query>"
  }
}