mem0ai/mem0 · error · Error

FastEmbed embed() returned no embeddings

Error message

FastEmbed embed() returned no embeddings

What it means

Thrown by FastEmbedEmbedder.embed() when iterating the fastembed output stream yields no defined first embedding for the single input text. FastEmbed returns results as async batches; if the stream completes without ever producing batch[0], the SDK cannot return a vector and raises rather than returning undefined. It indicates an anomalous library/model interaction, not bad input text.

Source

Thrown at mem0-ts/src/oss/src/embeddings/fastembed.ts:82

    return sdk.FlagEmbedding.init({ model: this.modelName });
  }

  private normalizeInput(text: string): string {
    return text.replace(/\n/g, " ");
  }

  async embed(text: string): Promise<number[]> {
    const normalizedText = this.normalizeInput(text);
    const model = await this.getEmbeddingModel();

    for await (const batch of model.embed([normalizedText])) {
      const embedding = batch[0];
      if (embedding !== undefined) {
        return embedding;
      }
    }

    throw new Error("FastEmbed embed() returned no embeddings");
  }

  async embedBatch(texts: string[]): Promise<number[][]> {
    const normalizedTexts = texts.map((text) => this.normalizeInput(text));
    const model = await this.getEmbeddingModel();
    const embeddings: number[][] = [];

    for await (const batch of model.embed(normalizedTexts)) {
      embeddings.push(...batch);
    }

    return embeddings;
  }
}

View on GitHub (pinned to 001c235229)

Solutions

1. Pin/align the fastembed version with what this mem0-ts release was tested against, then clear the fastembed model cache and let it re-download
2. Verify with a trivial input: await embedder.embed('hello') — if that also throws, the model artifacts are bad, not your text
3. Pre-trim inputs and avoid feeding whitespace-only strings

Example fix

// before
const vec = await embedder.embed(userText); // throws on empty stream

// after
const text = userText.trim();
if (!text) throw new Error('refusing to embed empty text');
const vec = await embedder.embed(text);

const text = raw.trim();
if (!text) throw new Error('refusing to embed empty text');
await embedder.embed(text);

try {
  vec = await embedder.embed(text);
} catch (e) {
  if (e instanceof Error && e.message === 'FastEmbed embed() returned no embeddings') {
    // library/model artifact problem: clear the fastembed cache and rebuild once
    throw new Error('FastEmbed produced no output — clear the ONNX model cache and retry init');
  }
  throw e;
}

When it happens

Trigger: A fastembed version whose embed() yields empty arrays for some inputs; model files corrupted after a partial download so inference returns nothing; an edge case where normalized input (e.g. whitespace-only text) produces no output row.

Common situations: After upgrading the optional fastembed peer dependency to an incompatible version; interrupted first-run model download leaving broken artifacts in the cache directory.

Related errors

• Unsupported FastEmbed model "${config.model}". Supported mod
• Error getting embedding from AWS Bedrock model ${this.model}
• AWS Bedrock model ${this.model} returned no embedding for on
• HuggingFace embed() returned no embeddings for model '${this
• LM Studio embedder failed: ${message}

AI-assisted analysis of mem0ai/mem0@001c235229 (2026-08-15). Data as JSON: /api/errors/26fe9c0c352eead1. Report an issue: GitHub.