mem0ai/mem0 · error · Error

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

Error message

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

What it means

search() in the Langchain adapter throws when the query vector's length differs from this.dimension. The query is passed straight to similaritySearchVectorWithScore, so a mismatched dimension would otherwise produce wrong results or an obscure store-specific error.

Source

Thrown at mem0-ts/src/oss/src/vector_stores/langchain.ts:112

      console.warn(
        "Langchain store might not support custom IDs on insert. Trying without IDs.",
        e,
      );
      await this.lcStore.addVectors(vectors, documents);
    }
  }

  async keywordSearch(): Promise<null> {
    return null;
  }

  async search(
    query: number[],
    topK: number = 5,
    filters?: SearchFilters, // filters parameter is received but will be ignored
  ): Promise<VectorStoreResult[]> {
    if (this.dimension && query.length !== this.dimension) {
      throw new Error(
        `Query vector dimension mismatch. Expected ${this.dimension}, got ${query.length}`,
      );
    }

    // --- Remove filter processing logic ---
    // Filters passed via mem0 interface are not reliably translatable to generic Langchain stores.
    // let lcFilter: any = undefined;
    // if (filters && ...) { ... }
    // console.warn("LangchainVectorStore: Passing filters directly..."); // Remove warning

    // Call similaritySearchVectorWithScore WITHOUT the filter argument
    const results = await this.lcStore.similaritySearchVectorWithScore(
      query,
      topK,
      // Do not pass lcFilter here
    );

    // Map Langchain results [Document, score] back to mem0 VectorStoreResult

View on GitHub (pinned to 001c235229)

Solutions

Re-embed queries with the same model used for inserts (use Memory.search rather than calling store.search with hand-made embeddings).
Update config.dimension to match the actual embedder output dimension.
If models were switched, reindex stored memories with the new embedder before searching.

Example fix

// before
const results = await store.search(staleQueryVector1536, 5); // dimension now 768

// after
const { embedding } = await memory.embedder.embed('query text');
const results = await store.search(embedding, 5);

Defensive patterns

Strategy: validation

Validate before calling

if (storeDimension && query.length !== storeDimension) {
  throw new Error(`Query is ${query.length}-d but store expects ${storeDimension}-d; re-embed the query`);
}

Try / catch

try {
  await store.search(query, topK);
} catch (e) {
  if (e instanceof Error && e.message.includes('Query vector dimension mismatch')) {
    // re-embed the query with the store's embedding model and retry
  }
  throw e;
}

Prevention

Prefer Memory.search() so queries are always embedded by the configured embedder.
Never reuse cached query embeddings across embedder changes.
Probe and pin the embedder dimension at startup.

When it happens

Trigger: Calling search() with a query embedded by a different model than the one whose dimension was configured/inferred — e.g. Memory's embedder changed after construction, or a manual search() call with a stale cached embedding.

Common situations: Swapping the embedding provider between add and search calls; using a persisted dimension from config while the live embedder differs; embedding the query with a different model version that changed dimensions (e.g. ada-002 1536 vs small model 1536 but MRL-truncated 256).

Related errors

AI-assisted analysis of mem0ai/mem0@001c235229 (2026-08-15). Data as JSON: /api/errors/ce6ab386b6a3580f. Report an issue: GitHub.