Search | Hivemind docs

Terminal

curl -X POST https://app.gethivemind.xyz/v3/search \
  -H "Authorization: Bearer $HIVEMIND_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "q": "how often can we deploy?",
    "budget": 2000,
    "targetTool": "claude-code"
  }'

json

{
  "payload": "--- RETRIEVED MEMORY CONTEXT ---\n[Context 1]: …",
  "memoryIds": ["…"],
  "candidateCount": 4,
  "droppedCount": 3,
  "tokensSaved": 148,
  "latencyMs": 287
}

payload is ready for a prompt. See Retrieval for what happens between the query and that string.

Parameters

	Default
`q`	required	The query
`budget`	`4000`	Hard token ceiling
`sourceRef`	none	Where the question is asked from, ranking origin, never a filter
`limit`	`10`	Cap on results. Recall keeps climbing past this, raise it deliberately, and watch the token count
`threshold`	calibrated	Relevance floor, 0–1
`searchMode`	`hybrid`	`memories`, `documents`, `hybrid`
`filters`	,	Metadata predicate
`rerank`	`true`	Cross-encoder re-scoring. Adds ~100ms; pass `false` if latency matters more than ordering
`rewriteQuery`	`false`	Also search alternative phrasings
`targetTool`	,	Shapes the payload for the receiver
`threadId`	,	Boost a thread
`include`	,	`memories`, `dropped`, `trace`

Filters

json

{
  "q": "database migrations",
  "filters": {
    "AND": [
      { "key": "team", "value": "platform" },
      { "key": "priority", "value": 5, "comparator": "gte", "numeric": true }
    ]
  }
}

Comparators: eq, ne, gt, gte, lt, lte, contains, in. Nest with AND and OR; negate inverts; numeric compares as numbers rather than text.

Empty results are normal

memoryIds is empty when nothing is relevant enough. Check its length rather than assuming there is always context to inject.