Tool reference · Maintenance & performance
memory_rebuild_embeddings
Recomputes the vectors used for semantic search for all active records, or only for one project or embedding space.
local stdio server writes destructive idempotent
Caution. Marked destructive: existing vectors are overwritten. Encoding every record is CPU-heavy and can take minutes on large stores.
When to use
- You switched the embedding model and old vectors no longer match.
- You changed the code embedder and only code records need refreshing.
- Semantic search misses records that keyword search finds.
Parameters
| Name | Type | Default | Description |
|---|---|---|---|
embedding_space | string | array | — | Optional: only re-encode rows in these spaces. |
project | string | — | — |
batch_size | integer | 32 | — |
limit | integer | — | — |
Example
Arguments
{
"embedding_space": "code",
"project": "billing-api",
"batch_size": 32
}Result shape
{
"rebuilt": 250,
"skipped": 0,
"embedding_space_filter": [
"code"
],
"project_filter": "billing-api"
}Use limit to test on a small batch first. skipped counts records whose encoding failed.
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:
re-encode every record (or every record in a given embedding space) and update the binary + float32 vectors. Idempotent. Pass embedding_space='code' to refresh only code rows after switching the code embedder. Returns {rebuilt: int, skipped: int}.