Tried to query vector index ${indexName} but this Memory ins

Error message

Tried to query vector index ${indexName} but this Memory instance doesn't have an attached vector db.

What it means

During recall's semantic search the embeddings are queried against a vector index, but the Memory instance has no vector store attached (this.vector undefined). The error is thrown defensively right before querying, naming the index that would have been used.

Source

Thrown at packages/memory/src/index.ts:738

          total: 0,
          page: page ?? 0,
          perPage: 0,
          hasMore: false,
        };
        span?.end({ output: { success: true }, attributes: { messageCount: 0 } });
        return result;
      }

      if (config?.semanticRecall && vectorSearchString && this.vector) {
        const result = await this.embedMessageContent(vectorSearchString!);
        usage = result.usage;
        const { embeddings, dimension } = result;
        const { indexName } = await this.createEmbeddingIndex(dimension, config);

        await Promise.all(
          embeddings.map(async embedding => {
            if (typeof this.vector === `undefined`) {
              throw new Error(
                `Tried to query vector index ${indexName} but this Memory instance doesn't have an attached vector db.`,
              );
            }

            const scopeFilter = resourceScope ? { resource_id: resourceId } : { thread_id: threadId };
            const userFilter = typeof config.semanticRecall === 'object' ? config.semanticRecall.filter : undefined;
            const combinedFilter = userFilter ? { $and: [scopeFilter, userFilter] } : scopeFilter;

            vectorResults.push(
              ...(await this.vector.query({
                indexName,
                queryVector: embedding,
                topK: vectorConfig.topK,
                filter: combinedFilter,
              })),
            );
          }),
        );

View on GitHub (pinned to 75dd419e61)

Solutions

Attach a vector store: pass `vector: new PGVector(...)`, `new FastVector(...)`, etc. when constructing Memory
Ensure the conditional that creates the vector store actually runs (check env vars/feature flags)
Disable semanticRecall if vector search is not intended

Example fix

// before
const memory = new Memory({ storage: store, embedder: embedder });
// after
import { PGVector } from '@mastra/pg';
const memory = new Memory({ storage: store, embedder: embedder, vector: new PGVector({ connectionString: process.env.DATABASE_URL }) });

Defensive patterns

Strategy: validation

Validate before calling

if (!memoryVector) throw new Error('Memory requires a vector store for semanticRecall; pass vector: new ...Vector(...)');

Type guard

function hasVectorStore(m: { vector?: unknown }): boolean {
  return m.vector !== undefined && m.vector !== null;
}

Prevention

Always construct Memory with a vector store when semanticRecall is enabled
Verify vector store construction errors/env vars at startup, not first recall
Add a startup assertion that memory.vector is defined when using semantic recall

When it happens

Trigger: Configuring semanticRecall without providing a `vector` option to `new Memory({...})`, then triggering a vector search; vector store constructor failing silently or being conditionally omitted (e.g. missing env vars for the provider).

Common situations: Copying Memory config examples without the vector field; vector store disabled when an API key env var is absent; switching storage backends that bundle a vector store to one that doesn't.

Related errors

AI-assisted analysis of mastra-ai/mastra@75dd419e61 (2026-08-30). Data as JSON: /api/errors/6885fa0a7302b4da. Report an issue: GitHub.