mem0ai/mem0 · error · Error

No predictions returned from Vertex AI

Error message

No predictions returned from Vertex AI

What it means

After a successful client.predict() call, embed() requires response.predictions to be a non-empty array. Vertex AI returned a response with no prediction records (or a shape without predictions), which the SDK treats as a protocol-level failure instead of indexing into undefined.

Source

Thrown at mem0-ts/src/oss/src/embeddings/vertexai.ts:178

      if (!(memoryAction in this.embeddingTypes)) {
        throw new Error(`Invalid memory action: ${memoryAction}`);
      }
      embeddingType = this.embeddingTypes[memoryAction];
    }

    const instance = this.formatInstance(text, embeddingType);
    const parameters = {
      outputDimensionality: this.embeddingDims,
    };

    const [response] = await this.client.predict({
      endpoint: this.endpoint(),
      instances: [this.helpers.toValue(instance) as any],
      parameters: this.helpers.toValue(parameters) as any,
    });

    if (!response.predictions || response.predictions.length === 0) {
      throw new Error("No predictions returned from Vertex AI");
    }

    const decoded = this.helpers.fromValue(response.predictions[0] as any);
    if (!isValidEmbedding(decoded)) {
      throw new Error("Failed to extract embedding values from response");
    }

    return decoded.embeddings.values;
  }

  async embedBatch(
    texts: string[],
    memoryAction: "add" | "update" | "search" = "add",
  ): Promise<number[][]> {
    if (!texts || texts.length === 0) {
      return [];
    }

View on GitHub (pinned to 001c235229)

Solutions

1. Retry once: transient empty responses from regional endpoints do occur
2. Verify the model is available and enabled in the exact region/location configured for the embedder
3. Reproduce with a direct curl/gcloud aiplatform predict call to see the raw response
4. Confirm you are on a current @google-cloud/aiplatform version and a stable (non-preview) model id

function isNoPredictionsError(err: unknown): boolean {
  return err instanceof Error && err.message === "No predictions returned from Vertex AI";
}

async function embedRetry(text: string, tries = 2) {
  for (let i = 0; ; i++) {
    try { return await embedder.embed(text); }
    catch (err) {
      if (i < tries && err instanceof Error && err.message === "No predictions returned from Vertex AI") continue;
      throw err;
    }
  }
}

When it happens

Trigger: Hitting a Vertex endpoint that returns an empty predictions list (model not enabled for the project, regional endpoint quirk, or a gateway/proxy stripping the field); passing a malformed instance that the service accepts but does not score; preview/new embedding models with a different response envelope.

Common situations: The aiplatform.googleapis.com embedding model is not enabled in the project/region; using a preview model id whose response shape differs; an intermediate proxy rewriting the gRPC/REST response.

Related errors

• No predictions returned from Vertex AI batch request
• Invalid memory action: ${memoryAction}
• Failed to extract embedding values from response
• Failed to extract embedding values from batch response
• Vertex AI embedBatch() returned ${allEmbeddings.length} embe

AI-assisted analysis of mem0ai/mem0@001c235229 (2026-08-15). Data as JSON: /api/errors/3ba7c76fc9e4a045. Report an issue: GitHub.