Mintplex-Labs/anything-llm · error · Error

Cohere Failed to embed: ${error}

Error message

Cohere Failed to embed: ${error}

What it means

Thrown after the embedding batches settle if any returned an error. Cohere per-batch errors are normalized into uniqueErrors entries shaped '[<error.type>]: <error.message>' and joined; the combined string is prefixed 'Cohere Failed to embed:'. Any failed batch aborts the whole call and no vectors are returned.

Source

Thrown at server/utils/EmbeddingEngines/cohere/index.js:90

        .filter((res) => !!res.error)
        .map((res) => res.error)
        .flat();

      if (errors.length > 0) {
        let uniqueErrors = new Set();
        errors.map((error) =>
          uniqueErrors.add(`[${error.type}]: ${error.message}`)
        );
        return { data: [], error: Array.from(uniqueErrors).join(", ") };
      }

      return {
        data: results.map((res) => res?.data || []).flat(),
        error: null,
      };
    });

    if (!!error) throw new Error(`Cohere Failed to embed: ${error}`);

    return data.length > 0 &&
      data.every((embd) => embd.hasOwnProperty("embedding"))
      ? data.map((embd) => embd.embedding)
      : null;
  }
}

module.exports = {
  CohereEmbedder,
};

View on GitHub (pinned to 526360e320)

Solutions

1. Parse the joined '[type]: message' tokens: 'auth'/'401' -> key; 'rate_limit'/'429' -> back off and reduce concurrency; 'request_too_large' -> shorten chunks.
2. Set EMBEDDING_MODEL_PREF to a Cohere embedding model (e.g. embed-english-v3.0, embed-multilingual-v3.0).
3. Reduce maxConcurrentChunks or batch size to stay under Cohere limits.
4. Retry after resolving the persistent cause; partial vectors are not returned.

// Pre-flight one small embed to validate key + model before bulk
try {
  await embedder.embedTextInput('ping');
} catch (e) {
  throw new Error(`Cohere embed preflight failed: ${e.message}`);
}

try {
  await embedder.embedChunks(chunks);
} catch (e) {
  const msg = e.message;
  if (/\[auth\]|401|unauthorized/i.test(msg)) rotateKey();
  else if (/rate|429|quota/i.test(msg)) reduceConcurrency();
  else if (/too.large|token/i.test(msg)) shortenChunks();
  else throw e;
}

When it happens

Trigger: Invalid/revoked COHERE_API_KEY surfacing as auth errors on the batch; 429 rate limits on embed-english-v3.0 or similar; input too long for the model's max token limit; embedding model id in EMBEDDING_MODEL_PREF not enabled for the account; network errors on some concurrent batches.

Common situations: Bulk-embedding a large corpus that trips Cohere rate limits; wrong model id (e.g. a rerank model used for embedding); oversized chunks; key rotated mid-job.

Related errors

• Azure OpenAI Failed to embed: ${error}
• Gemini Failed to embed: ${error}
• No Cohere API key was set.
• GenericOpenAI Failed to embed: ${error.message}
• Lemonade Failed to embed: [${error.type}]: ${error.message}

AI-assisted analysis of Mintplex-Labs/anything-llm@526360e320 (2026-08-13). Data as JSON: /api/errors/3918943333d9d0e3. Report an issue: GitHub.