Aiflow serves Gemini through Antigravity. You call it with the same OpenAI-compatible API and sk-... key as every other model. Aiflow converts the request to Gemini's format and converts the answer back, so you don't need a Google SDK or a Google API key.

Models

Model ID	Model
`ag/gemini-pro-agent`	Gemini 3.1 Pro
`ag/gemini-3.6-flash-high`	Gemini 3.6 Flash
`ag/gemini-3.7-flash-high`	Gemini 3.7 Flash
`ag/gemini-3.8-flash-high`	Gemini 3.8 Flash

Every ag/ model appears in /v1/models. Older antigravity-gemini-... IDs still work for existing clients.

Endpoints

Gemini models work on two endpoints:

Method	Path
`POST`	`/v1/chat/completions`
`POST`	`/v1/responses`

They are not available on /v1/messages, and Aiflow has no native Gemini generateContent endpoint.

Chat Completions

curl https://api.tryaiflow.io/v1/chat/completions \
  -H "Authorization: Bearer $AIFLOW_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ag/gemini-3.8-flash-high",
    "messages": [
      { "role": "system", "content": "Answer in one sentence." },
      { "role": "user", "content": "What is a mutex?" }
    ]
  }'

Responses

curl https://api.tryaiflow.io/v1/responses \
  -H "Authorization: Bearer $AIFLOW_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ag/gemini-pro-agent",
    "instructions": "Answer in one sentence.",
    "input": "What is a mutex?"
  }'

OpenAI SDK

import OpenAI from 'openai';

const client = new OpenAI({
  apiKey: process.env.AIFLOW_API_KEY,
  baseURL: 'https://api.tryaiflow.io/v1',
});

const completion = await client.chat.completions.create({
  model: 'ag/gemini-3.8-flash-high',
  messages: [{ role: 'user', content: 'Reply with ok' }],
});

console.log(completion.choices[0].message.content);

Supported input

Feature	Chat Completions	Responses
Text	`system`, `developer`, `user`, `assistant`	`instructions` and `input` messages
Images	`image_url` as a base64 data URL	`input_image` as a base64 data URL
Function tools	`tools` with `type: "function"`	`tools` with `type: "function"`
Tool results	`tool` messages	`function_call_output` items
Output limit	`max_tokens` or `max_completion_tokens`	`max_output_tokens`
Temperature	`temperature`	Not forwarded

Limits

Function tools only. Built-in tools such as web search and file search are rejected.
No previous_response_id. Send the full conversation in input on each request.
Buffered streaming. stream: true returns valid SSE, but the events arrive together after Gemini finishes the answer. Tokens are not streamed one by one.
Account-enabled keys only. A key that is limited to upstream routing gets 403. Contact support if you need Gemini access on your key.

Gemini usage counts against the same key budget as other models. Check it with the Usage API.

Gemini