Member Junction
    Preparing search index...
    • Reads a persisted embedding, preferring the binary column and falling back to the JSON column.

      The fallback also covers a binary value that fails validation, so a corrupt binary column never hides a good JSON copy. The two shapes differ (Float32Array vs number[]); both are accepted by SimpleVectorService.LoadVectors / AddVector / AddOrUpdateVector.

      Parameters

      • vectorBinary: string

        The binary column value (base64), or null when the row has none or it was not fetched.

      • vectorJSON: string

        The JSON column value, or null.

      Returns number[] | Float32Array<ArrayBufferLike>

      The vector, or null when neither column holds a usable vector.

      const vector = ReadStoredVector(note.EmbeddingVectorBinary, note.EmbeddingVector);
      if (vector) service.AddOrUpdateVector(note.ID, vector, metadata);