Tool reference · Save & recall
memory_search_fast
Runs the standard search with reranking and diversity turned off, so results are deterministic and no LLM is used.
local stdio server read-only
When to use
- You want predictable, repeatable results for the same query.
- You want to limit results to one kind of embedding (text, code, log or config).
- You call search from a hook or script where latency matters.
Parameters
| Name | Type | Default | Description |
|---|---|---|---|
queryrequired | string | — | — |
project | string | — | — |
type | "decision" | "fact" | "solution" | "lesson" | "convention" | "all" | "all" | — |
limit | integer | 10 | — |
detail | "compact" | "summary" | "full" | "full" | — |
branch | string | — | — |
fusion | "rrf" | "legacy" | "rrf" | — |
embedding_space | string | array | — | Filter to one or more embedding spaces (text|code|log|config). |
Example
Arguments
{
"query": "staging database port",
"project": "my-api",
"limit": 3,
"detail": "compact"
}Result shape
{
"query": "staging database port",
"total": 1,
"detail": "compact",
"fusion": "rrf",
"total_tokens": 41,
"results": {
"fact": [
{
"id": 1843,
"type": "fact",
"title": "Staging database runs on port 5433; production uses 5432.",
"project": "my-api",
"score": 0.874,
"importance": "medium",
"created_at": "2026-09-20T10:20:11Z",
"_tokens": 41
}
]
},
"semantic_diagnostics": [],
"mode": "fast"
}Values are illustrative; the keys follow the server's handler. MCP clients receive the result as JSON text content.
Server description
The description the server sends to your agent in tools/list, captured from the v14.7.0 source:
like memory_recall but with rerank=False, diverse=False forced. Deterministic fast path — zero LLM, FastEmbed-only.