Tool reference · Save & recall
memory_answer
Writes an answer to a question using only records from one project, cites the exact quotes it relied on, and checks the answer with a second LLM pass.
local stdio server writes
Caution. Requires a configured reasoning LLM provider and an explicit project. One call can make up to eight LLM requests.
When to use
- You need a direct answer with citations, not a list of search hits.
- Records may disagree over time and you want the latest value, with the replaced value named.
- You want the tool to say "insufficient" rather than guess when memory does not contain the answer.
Parameters
| Name | Type | Default | Description |
|---|---|---|---|
queryrequired | string | — | (min length 1) |
projectrequired | string | — | (min length 1) |
branch | string | — | — |
type | "all" | "fact" | "decision" | "solution" | "lesson" | "convention" | — | — |
limit | integer | 10 | (min 1, max 20) |
max_bytes | integer | 24000 | (min 512, max 24000) |
followup | boolean | true | — |
Example
Arguments
{
"query": "What does the nightly billing job use for locking?",
"project": "my-api",
"limit": 10
}Result shape
{
"answer": "It uses PostgreSQL advisory locks; this replaced the earlier Redis lock.",
"draft": {
"status": "supported",
"answer": "It uses PostgreSQL advisory locks.",
"claims": [
{
"subject": "nightly billing job",
"event": "uses PostgreSQL advisory locks",
"time": "2026-09-20",
"modality": "asserted",
"citations": [
{
"source_id": 1842,
"quote": "Use PostgreSQL advisory locks",
"start": 0,
"end": 29
}
]
}
],
"missing": null,
"rejection": null
},
"verification": {
"supported": true,
"reason": "The cited record states the current lock mechanism."
},
"search_calls": 2,
"followup_query": null,
"evidence": [
{
"id": 1842,
"type": "decision",
"project": "my-api",
"content": "Use PostgreSQL advisory locks…",
"created_at": "2026-09-20T10:14:03Z"
}
],
"negative": null,
"caveat": null
}Citation start/end offsets refer to the content of the returned evidence items.
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:
Generate and verify a cited answer using the configured reasoning LLM. Evidence carries recording dates; when records about the same subject disagree, the latest one gives the current value and the answer names the value it replaced. First runs negative retrieval: a contradiction-seeking second search; a score >= 0.60 hands both sides to the reader and answers with a caveat (MEMORY_CONTRADICTION_POLICY=abstain refuses instead), 0.30-0.60 answers with a caveat (see `negative`); MEMORY_CONTRADICTION_SCORER=jev scores the pairs with TypeSafe's Jev instead of the LLM. Up to one missing-relation retrieval and eight LLM calls including inversion retry and bounded quote repair. Explicit project required. Citation offsets refer to returned evidence content. Ordinary recall remains local.