Reindex API

The reindex API manages embedding backfill for memories that were imported or created without vectors.

Why it exists

MindStore can accept imports even when embeddings are skipped:

• the import was too large to embed inline
• no embedding provider was configured at import time
• a provider call failed transiently

Instead of silently leaving those memories half-indexed forever, MindStore now records an indexing job and exposes its status.

Endpoints

GET /api/v1/reindex

Returns the current embedding state for the authenticated user.

Response shape:

{
  "total": 120,
  "withEmbeddings": 100,
  "withoutEmbeddings": 20,
  "needsReindex": true,
  "latestJob": {
    "id": "uuid",
    "status": "pending",
    "remainingCount": 20
  }
}

POST /api/v1/reindex

Runs one embedding-backfill batch immediately by default.

Optional request body:

{
  "batchSize": 50,
  "runNow": true
}

Set "runNow": false to only queue a backfill job without processing a batch immediately.

Possible outcomes:

• pending: more memories still need embeddings
• completed: all memories now have embeddings
• blocked: no embedding provider is configured yet
• failed: the provider call failed and should be retried

Operator workflow

For private installs or VPS jobs, you can process queued indexing batches with:

npm run jobs:run-indexing

This is intended for scheduled or repeated execution, similar to the plugin-jobs runner.