Gemini API

The Gemini endpoint exposes the Google Gemini API through Neon AI Gateway. Use it when you're already working with the google-genai SDK and want to keep your existing code. For most use cases, the chat completions endpoint is simpler to set up and works with Gemini models via the OpenAI SDK.

Supported actions: :generateContent and :streamGenerateContent

Endpoint pattern: https://<branch-host>/ai-gateway/gemini/v1beta/models/<model>:<action>

Supported models

This endpoint accepts Google models only:

Sending a non-Google model ID returns 400 model is not available on this endpoint. Use the chat completions endpoint if you want to call Gemini models alongside other providers from the same code.

Basic request

import { GoogleGenAI } from '@google/genai';

const client = new GoogleGenAI({
  apiKey: process.env.NEON_AI_GATEWAY_TOKEN,
  httpOptions: {
    baseUrl: `${process.env.NEON_AI_GATEWAY_BASE_URL}/ai-gateway/gemini`,
  },
});

const response = await client.models.generateContent({
  model: 'gemini-2-5-flash',
  contents: [{ role: 'user', parts: [{ text: 'What is Neon?' }] }],
});

console.log(response.text);

Streaming

const stream = await client.models.generateContentStream({
  model: 'gemini-2-5-flash',
  contents: [{ role: 'user', parts: [{ text: 'Explain database branching.' }] }],
});

for await (const chunk of stream) {
  process.stdout.write(chunk.text ?? '');
}

URL structure

The gateway uses the model ID and action directly in the URL path. The google-genai SDK constructs this automatically from the base URL and model parameter:

base_url: https://<branch-host>/ai-gateway/gemini
model:    gemini-2-5-flash
action:   generateContent or streamGenerateContent

→ https://<branch-host>/ai-gateway/gemini/v1beta/models/gemini-2-5-flash:generateContent
→ https://<branch-host>/ai-gateway/gemini/v1beta/models/gemini-2-5-flash:streamGenerateContent

When calling the REST API directly, the model ID and action must appear in the path as shown above.

Error handling

For authentication, quota, and upstream errors, see Troubleshooting.

Next steps

• Models: full model catalog
• Chat completions: use Gemini models via the unified OpenAI-compatible endpoint
• Authentication: credential scopes and branch binding