rohitg00/agentmemory · error · Error

Gemini embedding failed (${response.status}): ${err}

Error message

Gemini embedding failed (${response.status}): ${err}

What it means

`embedBatch` in the Gemini provider throws when Google's embedding endpoint returns a non-ok HTTP status, including the status code and raw response body in the message. The key was present (constructor passed) but the API call itself failed — auth, quota, model, or request-shape problems.

Source

Thrown at src/providers/embedding/gemini.ts:43

    const results: Float32Array[] = [];

    for (let i = 0; i < texts.length; i += BATCH_LIMIT) {
      const chunk = texts.slice(i, i + BATCH_LIMIT);
      const response = await fetchWithTimeout(`${API_BASE}?key=${this.apiKey}`, {
        method: "POST",
        headers: { "Content-Type": "application/json" },
        body: JSON.stringify({
          requests: chunk.map((t) => ({
            model: MODEL,
            content: { parts: [{ text: t }] },
            outputDimensionality: this.dimensions,
          })),
        }),
      });

      if (!response.ok) {
        const err = await response.text();
        throw new Error(`Gemini embedding failed (${response.status}): ${err}`);
      }

      const data = (await response.json()) as {
        embeddings: Array<{ values: number[] }>;
      };

      for (const emb of data.embeddings) {
        results.push(l2Normalize(new Float32Array(emb.values)));
      }
    }

    return results;
  }
}

let zeroNormWarned = false;

function l2Normalize(vec: Float32Array): Float32Array {

View on GitHub (pinned to e04ba88819)

Solutions

1. Inspect the response body embedded in the error — Google's message names the exact cause (API key invalid, quota exceeded, model not found).
2. If 401/403: generate a new key at Google AI Studio and ensure the Generative Language API is enabled/not restricted for it.
3. If 429: implement backoff-and-retry, lower request rate and batch size, or upgrade from the free tier.
4. If 400: verify the model name is a valid embedding model and texts are non-empty and within token limits.
5. If 5xx: retry with exponential backoff; check Google's status dashboard if it persists.

Example fix

// before
const vec = await provider.embed(text); // throws on 429 with body
// after
for (let attempt = 0; attempt < 3; attempt++) {
  try { return await provider.embed(text); }
  catch (e) {
    if (!/\(429\)|\(5\d\d\)/.test(e.message) || attempt === 2) throw e;
    await sleep(2 ** attempt * 500);
  }
}

// Preflight the key against a cheap endpoint before embedding batches
const res = await fetch(`https://generativelanguage.googleapis.com/v1/models?key=${process.env.GEMINI_API_KEY}`);
if (!res.ok) throw new Error(`GEMINI_API_KEY preflight failed: ${res.status}`);

async function embedWithRetry(text: string, tries = 3) {
  for (let i = 0; i < tries; i++) {
    try { return await provider.embed(text); }
    catch (e) {
      const m = e instanceof Error ? e.message : String(e);
      const status = Number(/Gemini embedding failed \((\d+)\)/.exec(m)?.[1]);
      if (status === 401 || status === 403 || (status >= 400 && status < 500 && status !== 429)) throw e;
      await new Promise(r => setTimeout(r, 2 ** i * 500 + Math.random() * 250));
    }
  }
  throw new Error("Gemini embed: retries exhausted");
}

When it happens

Trigger: Any `embed()`/`embedBatch()` call where the Gemini embeddings endpoint responds 400 (malformed request, unsupported model like a non-embedding model name), 401/403 (bad/missing permissions on the key), 429 (quota/rate limit), or 5xx (Google-side outage).

Common situations: Key restricted by API restrictions in Google Cloud console (Generative Language API not allowed); free-tier per-minute quota exhausted under batch load; using `models/embedding-001` vs newer `gemini-embedding` naming mismatch; request disabled via API key restrictions or region not enabled.

Related errors

• Cohere embedding failed (${response.status}): ${err}
• POST ${url} failed: ${res.status} ${res.statusText}${suffix}
• ${init?.method || "GET"} ${path} -> ${res.status} ${res.stat
• MiniMax API error ${response.status}: ${text}
• OpenAI API error (${response.status}): ${text}

AI-assisted analysis of rohitg00/agentmemory@e04ba88819 (2026-08-30). Data as JSON: /api/errors/ef70433c9a2acda4. Report an issue: GitHub.