Gemini Native Format

Native Google Gemini API protocol for direct SDK integration

Endpoints

Generate Content

POST http://127.0.0.1:8045/v1beta/models/{model}:generateContent

Stream Generate Content

POST http://127.0.0.1:8045/v1beta/models/{model}:streamGenerateContent?alt=sse

List Models

GET http://127.0.0.1:8045/v1beta/models

Get Model Info

GET http://127.0.0.1:8045/v1beta/models/{model}

Count Tokens

POST http://127.0.0.1:8045/v1beta/models/{model}:countTokens

The Gemini native protocol endpoint provides direct compatibility with Google AI SDKs and supports advanced features like thinking signatures, grounding, and tool calling.

Authentication

Authorization · string · required

Bearer token authentication

Authorization: Bearer sk-antigravity

Alternatively, use query parameter:

key · string

API key as query parameter

?key=sk-antigravity

Path Parameters

model · string · required

Model name:

• gemini-3-flash - Fast and efficient
• gemini-3-pro-high - High quality reasoning
• gemini-3-pro-low - Cost-efficient
• gemini-3-pro-image - Image generation
• claude-sonnet-4-6 - Mapped Claude model

Request Headers

Content-Type · string · required

Must be application/json

Request Body

contents · array · required

Conversation content in Gemini format

Show Content Object

role · string · required

Message role: user or model

parts · array · required

Message parts (text, images, function calls)

Show Part Types

text · string

Text content

thought · boolean

Mark as thinking content

thoughtSignature · string

Cryptographic signature for thought validation

inlineData · object

Embedded media data

Show Inline Data

mimeType · string

MIME type, e.g., image/jpeg, audio/mp3

data · string

Base64-encoded data

functionCall · object

Function/tool call

Show Function Call

name · string

Function name

args · object

Function arguments as JSON object

functionResponse · object

Function execution result

Show Function Response

name · string

Function name

response · object

Result data

generationConfig · object

Generation parameters

Show Generation Config

maxOutputTokens · integer

Maximum output tokens (automatically capped to model limits)

temperature · number

Sampling temperature (0.0 to 2.0)

topP · number

Nucleus sampling (0.0 to 1.0)

topK · integer

Top-K sampling

stopSequences · array

Stop sequences for generation

thinkingConfig · object

Extended thinking configuration

Show Thinking Config

thinkingBudget · integer

Token budget for reasoning (e.g., 8192, 16384, 24576)

thinkingLevel · string

Thinking level: NONE, LOW, MEDIUM, HIGH (Auto-converted to budget internally)

tools · array

Available tools/functions

Show Tool Definition

functionDeclarations · array

Function declarations

Show Function

name · string · required

Function name

description · string

Function description

parameters · object

JSON Schema for parameters

googleSearch · object

Enable Google Search grounding (automatic web search)

systemInstruction · object

System-level instruction

Show System Instruction

parts · array

Instruction parts (same structure as content parts)

project · string

Project ID (automatically injected by proxy)

Response Format

Non-Streaming Response

candidates · array

Generated candidates

Show Candidate Object

content · object

Generated content

Show Content

role · string

Always model

parts · array

Response parts (text, function calls, thinking)

finishReason · string

Completion reason: STOP, MAX_TOKENS, SAFETY, RECITATION

index · integer

Candidate index

groundingMetadata · object

Search grounding results

Show Grounding Metadata

webSearchQueries · array

Search queries executed

groundingChunks · array

Web sources with titles and URIs

usageMetadata · object

Token usage statistics

Show Usage Metadata

promptTokenCount · integer

Input tokens

candidatesTokenCount · integer

Output tokens

totalTokenCount · integer

Total tokens

cachedContentTokenCount · integer

Cached tokens (context caching)

modelVersion · string

Actual model version used

Example: Basic Generation

curl -X POST "http://127.0.0.1:8045/v1beta/models/gemini-3-flash:generateContent" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-antigravity" \
  -d '{
    "contents": [
      {
        "role": "user",
        "parts": [{"text": "Explain how AI works"}]
      }
    ]
  }'

Example: Streaming

curl -X POST "http://127.0.0.1:8045/v1beta/models/gemini-3-pro-high:streamGenerateContent?alt=sse" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-antigravity" \
  -d '{
    "contents": [
      {
        "role": "user",
        "parts": [{"text": "Write a story about space exploration"}]
      }
    ],
    "generationConfig": {
      "temperature": 0.8,
      "maxOutputTokens": 2048
    }
  }'

Example: With System Instruction

curl -X POST "http://127.0.0.1:8045/v1beta/models/gemini-3-flash:generateContent" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-antigravity" \
  -d '{
    "systemInstruction": {
      "parts": [{"text": "You are a helpful coding tutor. Always provide examples."}]
    },
    "contents": [
      {
        "role": "user",
        "parts": [{"text": "How do I sort a list in Python?"}]
      }
    ]
  }'

Example: Function Calling

curl -X POST "http://127.0.0.1:8045/v1beta/models/gemini-3-flash:generateContent" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-antigravity" \
  -d '{
    "contents": [
      {
        "role": "user",
        "parts": [{"text": "What is the weather in Paris?"}]
      }
    ],
    "tools": [
      {
        "functionDeclarations": [
          {
            "name": "get_weather",
            "description": "Get current weather for a city",
            "parameters": {
              "type": "object",
              "properties": {
                "city": {"type": "string"}
              },
              "required": ["city"]
            }
          }
        ]
      }
    ]
  }'

Example: Google Search Grounding

curl -X POST "http://127.0.0.1:8045/v1beta/models/gemini-3-pro-high:generateContent" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-antigravity" \
  -d '{
    "contents": [
      {
        "role": "user",
        "parts": [{"text": "What are the latest developments in quantum computing?"}]
      }
    ],
    "tools": [
      {"googleSearch": {}}
    ]
  }'

Example: Multi-Modal (Image)

curl -X POST "http://127.0.0.1:8045/v1beta/models/gemini-3-flash:generateContent" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-antigravity" \
  -d '{
    "contents": [
      {
        "role": "user",
        "parts": [
          {"text": "Describe this image:"},
          {
            "inlineData": {
              "mimeType": "image/jpeg",
              "data": "base64-encoded-image-data"
            }
          }
        ]
      }
    ]
  }'

Example: Extended Thinking

curl -X POST "http://127.0.0.1:8045/v1beta/models/gemini-3-pro-high:generateContent" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-antigravity" \
  -d '{
    "contents": [
      {
        "role": "user",
        "parts": [{"text": "Solve this complex math problem step by step: ..."}]
      }
    ],
    "generationConfig": {
      "thinkingConfig": {
        "thinkingBudget": 16384
      }
    }
  }'

Model Information

Get detailed model specs:

curl "http://127.0.0.1:8045/v1beta/models/gemini-3-pro-high" \
  -H "Authorization: Bearer sk-antigravity"

Response includes:

• maxOutputTokens - Maximum output limit
• supportsThinking - Extended thinking support
• supportedGenerationMethods - Available methods
• inputTokenLimit - Context window size

Token Counting

curl -X POST "http://127.0.0.1:8045/v1beta/models/gemini-3-flash:countTokens" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-antigravity" \
  -d '{
    "contents": [
      {
        "role": "user",
        "parts": [{"text": "Count tokens in this message"}]
      }
    ]
  }'

Response:

{
  "totalTokens": 8
}

Features

• Native Gemini Protocol: Full compatibility with Google AI SDK
• Auto Project Injection: Automatically injects project ID from account pool
• Dynamic Token Limits: Automatically caps maxOutputTokens to model limits
• Thinking Signatures: Validates and preserves thinking signatures across turns
• MCP Fuzzy Matching: Intelligent tool name matching for MCP servers
• Grounding Support: Google Search integration for factual queries
• Context Caching: Automatic caching for repeated contexts

Error Format

{
  "error": {
    "code": 400,
    "message": "Invalid model name",
    "status": "INVALID_ARGUMENT"
  }
}

Common error codes:

• 400 - Invalid request
• 401 - Authentication failed
• 429 - Quota exceeded (triggers auto-retry)
• 500 - Internal error