can1357/oh-my-pi · error · SearchProviderError

Gemini Developer API error

Error message

Gemini Developer API error (${response.status}): ${errorText}

What it means

Thrown by callGeminiDeveloperSearch in the Gemini web-search provider when the Gemini Developer API (streamGenerateContent endpoint) returns a non-OK HTTP status. Before throwing, the provider redacts the API key from the response body and attempts to classify the error via classifyProviderHttpError; this generic SearchProviderError is only the fallback for statuses/text that classifier does not recognize. The HTTP status and raw (redacted) error body are embedded in the message so the developer can see Google's actual reason.

Solutions

1. Read the HTTP status and errorText embedded in the message; fix the specific cause (enable Generative Language API for the key, correct model id, fix request params).
2. Verify GEMINI_API_KEY (or provider 'google' key) is a valid Developer API key at https://aistudio.google.com/apikey and not IP/referrer-restricted.
3. If 429/401/403 quota-or-auth style, rely on the classifier's specific error or rotate to a different credential/provider.
4. Retry later on 5xx; the request already went through fetchWithRetry with exponential backoff, so persistent 5xx means a Google-side incident or a persistent request problem.

Example fix

// before: key from Vertex AI service account JSON used against developer endpoint
const apiKey = process.env.VERTEX_KEY;
// after: dedicated AI Studio developer API key
const apiKey = process.env.GEMINI_API_KEY; // https://aistudio.google.com/apikey

if (!process.env.GEMINI_API_KEY) throw new Error('Set GEMINI_API_KEY before using the Gemini search provider');

import { SearchProviderError } from './search-provider-error';
function isSearchProviderError(e: unknown): e is SearchProviderError {
  return e instanceof SearchProviderError && e.provider === 'gemini';
}

try {
  const result = await searchGemini(params);
} catch (err) {
  if (err instanceof SearchProviderError && err.provider === 'gemini') {
    logger.error('Gemini search failed', { status: err.statusCode, message: err.message });
    if (err.statusCode && err.statusCode >= 500) return retryWithBackoff();
    return null; // 4xx: do not retry
  }
  throw err;
}

When it happens

Trigger: Any non-2xx response from POST {endpoint}/models/{model}:streamGenerateContent?alt=sse whose status/body is not matched by classifyProviderHttpError — e.g. 400 invalid request body, 403 API key restrictions or disabled API, 404 unknown model id on the developer API endpoint, 5xx outages.

Common situations: Using an API key created for Vertex AI instead of the Developer API; Google Cloud APIs (Generative Language API) not enabled for the key's project; API key restricted by referrer/IP or to other APIs; requesting a model name not available on the developer endpoint; transient Google 5xx incidents.

Related errors

• Gemini Files API delete failed with HTTP
• Gemini Files API upload finalization failed with HTTP
• Jina API error ( )
• Gemini image request failed
• Anthropic API error ( )

AI-assisted analysis of can1357/oh-my-pi@9690622007 (2026-08-31). Data as JSON: /api/errors/0ff9e5caa9fbf3cc. Report an issue: GitHub.

Appendix: source

Thrown at packages/coding-agent/src/web/search/providers/gemini.ts:597

				? { "cf-aig-authorization": `Bearer ${apiKey}` }
				: { "x-goog-api-key": apiKey }),
			"Content-Type": "application/json",
			Accept: "text/event-stream",
		},
		body: JSON.stringify(requestBody),
		signal: withHardTimeout(signal, timeoutMs),
		fetch: fetchImpl,
		maxAttempts: MAX_RETRIES + 1,
		defaultDelayMs: attempt => BASE_DELAY_MS * 2 ** attempt,
		maxDelayMs: RATE_LIMIT_BUDGET_MS,
	});

	if (!response.ok) {
		const rawErrorText = await response.text();
		const errorText = apiKey ? rawErrorText.split(apiKey).join("[redacted]") : rawErrorText;
		const classified = classifyProviderHttpError("gemini", response.status, errorText);
		if (classified) throw classified;
		throw new SearchProviderError(
			"gemini",
			`Gemini Developer API error (${response.status}): ${errorText}`,
			response.status,
		);
	}

	if (!response.body) {
		throw new SearchProviderError("gemini", "Gemini API returned no response body", 500);
	}

	return finalizeGeminiSearchResult(await parseGeminiSearchStream(response.body, model), fetchImpl, signal);
}

/**
 * Executes a web search using Google Gemini with Google Search grounding.
 */
export async function searchGemini(params: GeminiSearchParams): Promise<SearchResponse> {
	const selectedModel = resolveGeminiSearchModel(params.geminiModel);

View on GitHub (pinned to 9690622007)