continuedev/continue · error · Error

data.error.message

Error message

data.error.message

What it means

Thrown by processGeminiResponse when the Gemini API response JSON contains an 'error' field; the error message from the payload is re-thrown directly. This surfaces Google's own error text (e.g. API key issues, safety blocks, quota, invalid model) during chat streaming.

Source

Thrown at core/llm/llms/Gemini.ts:400

      if (buffer.startsWith(",")) {
        buffer = buffer.slice(1);
      }

      const parts = buffer.split("\n,");

      let foundIncomplete = false;
      for (let i = 0; i < parts.length; i++) {
        const part = parts[i];
        let data: GeminiChatResponse;
        try {
          data = JSON.parse(part) as GeminiChatResponse;
        } catch (e) {
          foundIncomplete = true;
          continue; // yo!
        }

        if ("error" in data) {
          throw new Error(data.error.message);
        }

        // In case of max tokens reached, gemini will sometimes return content with no parts, even though that doesn't match the API spec
        const contentParts = data?.candidates?.[0]?.content?.parts;
        if (contentParts) {
          const textParts: MessagePart[] = [];
          const toolCalls: ToolCallDelta[] = [];

          for (const part of contentParts) {
            if ("text" in part) {
              textParts.push({ type: "text", text: part.text });
            } else if ("functionCall" in part) {
              const thoughtSignature = part.thoughtSignature;
              toolCalls.push({
                type: "function",
                id: part.functionCall.id ?? uuidv4(),
                function: {
                  name: part.functionCall.name,

View on GitHub (pinned to 5522c6f44c)

Solutions

  1. Match the message text: 'API key not valid' -> fix GEMINI_API_KEY; RESOURCE_EXHAUSTED -> backoff/quota; 'not found' -> fix model name
  2. Test the key with a curl call to generativelanguage.googleapis.com
  3. Add exponential backoff for RESOURCE_EXHAUSTED quota messages
  4. Use the current model name (e.g. gemini-1.5-pro/gemini-2.0-flash) from Google's docs

Example fix

// before
new Gemini({ apiKey: process.env.GEMINI_API_KEY, model: 'gemini-pro' });
// after
new Gemini({ apiKey: process.env.GEMINI_API_KEY?.trim(), model: 'gemini-2.0-flash' });
Defensive patterns

Strategy: try-catch

Validate before calling

const key = process.env.GEMINI_API_KEY?.trim();
if (!key) throw new Error('GEMINI_API_KEY required');

Type guard

function isGeminiApiError(e: unknown): boolean {
  return e instanceof Error && /API key|quota|safety|RESOURCE_EXHAUSTED|PERMISSION|not found/i.test(e.message);
}

Try / catch

try { for await (const m of geminiStream) { /* ... */ } }
catch (e) {
  if (isGeminiApiError(e)) { handleConfigOrQuota(e); return; }
  throw e;
}

Prevention

When it happens

Trigger: Streaming chat with an invalid GEMINI_API_KEY (API key not valid), quota exhaustion (429 RESOURCE_EXHAUSTED), requesting a nonexistent/deprecated model, or region-blocked API access, each returned as an error object in the response body.

Common situations: Expired/truncated GEMINI_API_KEY, free-tier keys hitting RPM/TPM limits during indexing loops, using gemini-pro after deprecation, or calling from an unsupported region/VPN.

Related errors

AI-assisted analysis of continuedev/continue@5522c6f44c (2026-08-27). Data as JSON: /api/errors/83e626cf6353e5e6. Report an issue: GitHub.