mem0ai/mem0 · error · Error

Query dimension mismatch. Expected ${this.dimension}, got ${

Error message

Query dimension mismatch. Expected ${this.dimension}, got ${query.length}

What it means

search() rejects any query vector whose length differs from the store's configured dimension before scanning rows, because cosine/dot comparison between differently-sized Float32 arrays is meaningless (and would produce NaN scores). The dimension is fixed when the collection is created, so a mismatch indicates the query was embedded with a different model than the stored data.

Source

Thrown at mem0-ts/src/oss/src/vector_stores/memory.ts:358

          id: s.id,
          payload: s.payload,
          score: s.score,
        }));

      return results;
    } catch (error) {
      console.error("Error during keyword search:", error);
      return null;
    }
  }

  async search(
    query: number[],
    topK: number = 10,
    filters?: SearchFilters,
  ): Promise<VectorStoreResult[]> {
    if (query.length !== this.dimension) {
      throw new Error(
        `Query dimension mismatch. Expected ${this.dimension}, got ${query.length}`,
      );
    }

    const rows = this.db.prepare(`SELECT * FROM vectors`).all() as any[];
    const results: VectorStoreResult[] = [];

    for (const row of rows) {
      const vector = new Float32Array(
        row.vector.buffer,
        row.vector.byteOffset,
        row.vector.byteLength / 4,
      );
      const payload = this.normalizePayload(JSON.parse(row.payload));
      const memoryVector: MemoryVector = {
        id: row.id,
        vector: Array.from(vector),
        payload,

View on GitHub (pinned to 001c235229)

Solutions

Use the exact same embedding model/config for search queries as for add()/insert() — ideally route both through the same Memory instance.
If the stored data is from an old model, wipe the DB and re-embed with the new model.
In tests, pass a consistent fake embedder (e.g. deterministic N-dim) to both the store config and query embedding.
Check query.length against store.dimension before calling search() in generic pipeline code.

Example fix

// before
const results = await store.search(otherEmbedder.embed('hello'), 5); // 384 vs 1536 -> throws

// after
const results = await store.search(memoryEmbedder.embed('hello'), 5); // same model as insert

Defensive patterns

Strategy: validation

Validate before calling

const expected = store.dimension ?? (await embedder.embed('probe')).length;
if (query.length !== expected) {
  throw new Error(`Query dim ${query.length} != store dim ${expected}; check embedder config`);
}

Type guard

const matchesStoreDim = (q: number[], dim: number): boolean => Array.isArray(q) && q.length === dim;

Try / catch

try { results = await store.search(query, topK, filters); }
catch (e) {
  if (e instanceof Error && e.message.startsWith('Query dimension mismatch')) {
    // re-embed query with the same model used for inserts, then retry
  } else throw e;
}

Prevention

Route add() and search() through the same Memory/embedder instance.
In tests, use a deterministic fake embedder with fixed output length everywhere.
Never hand-craft query vectors; always embed.

When it happens

Trigger: Embedding the search query with a different provider/model than the one used for inserts (e.g. store built with 1536-dim OpenAI vectors, query embedded with a 384-dim local model); calling search() with a raw hand-made vector of arbitrary length; environment-dependent embedder defaults (prod vs test).

Common situations: Swapping the embedder config after data was already stored; unit tests using fake embeddings of length N while the store was created with M; copy-pasting a query pipeline that uses a different embedder instance than the memory instance.

Related errors

AI-assisted analysis of mem0ai/mem0@001c235229 (2026-08-15). Data as JSON: /api/errors/37c9d6fb7c29c0f1. Report an issue: GitHub.