Cherry Studio integration

Overview

Cherry Studio is a powerful AI chat client that supports multiple providers and advanced features. Antigravity Manager provides seamless integration with native support for web search citations and streaming responses.

Antigravity v4.1.21 introduced comprehensive Cherry Studio support, including SSE event completion and web search citation display.

Prerequisites

Antigravity Manager installed and running
Cherry Studio installed (download here)
At least one active account

Configuration

Start Antigravity proxy

Open Antigravity Manager
Navigate to API Proxy tab
Enable the proxy server
Verify port (default: 8045)

Add provider in Cherry Studio

Open Cherry Studio settings
Go to Providers or Models
Add new custom provider:
- Name: Antigravity
- Base URL: http://127.0.0.1:8045/v1
- API Key: sk-antigravity
- Protocol: OpenAI

Configure models

Add models you want to use:

claude-sonnet-4-6
claude-sonnet-4-6-thinking
gemini-3-flash
gemini-3-pro-high
gemini-3-pro-image (for image generation)

Supported features

Chat completions

Basic chat works out of the box:

{
  "model": "claude-sonnet-4-6",
  "messages": [
    {"role": "user", "content": "Hello!"}
  ]
}

Antigravity automatically:

Routes to available accounts
Handles streaming responses
Manages quota protection

Extended thinking

Use thinking models for complex reasoning:

Adaptive thinking
Fixed budget

{
  "model": "claude-sonnet-4-6-thinking",
  "thinking": {
    "type": "adaptive"
  }
}

Antigravity maps this to 24576 token thinking budget.

{
  "model": "claude-sonnet-4-6-thinking",
  "thinking": {
    "budget": 16384
  }
}

Explicitly set thinking token budget.

Antigravity v4.1.21+ automatically converts Cherry Studio’s thinking.type: "adaptive" to a fixed budget compatible with Gemini’s API.

Web search with citations

Cherry Studio’s web search feature works seamlessly:

Enable web search in Cherry Studio chat
Antigravity automatically:
- Injects Google Search tool
- Executes search queries
- Returns results with citations
- Displays source links natively

Example output:

Based on my research:

**Source 1**: [Google Gemini API](https://ai.google.dev/api)
- Gemini 3 supports extended thinking mode

**Source 2**: [Anthropic Documentation](https://docs.anthropic.com)
- Claude models support adaptive reasoning

Cherry Studio will display these citations as clickable links.

Image generation

Configure image generation models with custom parameters:

Add image model

In Cherry Studio, add model: gemini-3-pro-image

Configure image settings

Go to model settings in Cherry Studio:

size

string

Image dimensions, e.g., 1920x1080, 1024x1024, 16:9Default: 1024x1024

quality

string

Quality level: standard, medium, hdMapping:

standard → 1K resolution
medium → 2K resolution
hd → 4K resolution

number

Number of images to generate (1-10)Default: 1

Generate images

Simply describe the image in chat:

Generate: A futuristic cityscape at sunset

Antigravity will:

Route to image generation endpoint
Apply configured size/quality
Return image as Markdown

Cherry Studio supports image input:

Click image attachment icon
Select image file
Add text prompt
Send message

Antigravity handles:

Image encoding (Base64)
Multi-modal request formatting
Protocol conversion to Gemini

Advanced configuration

Model-specific settings

Configure per-model parameters in Cherry Studio:

Chat model
Thinking model
Image model

{
  "model": "claude-sonnet-4-6",
  "temperature": 0.7,
  "max_tokens": 4096,
  "top_p": 0.9
}

{
  "model": "claude-sonnet-4-6-thinking",
  "temperature": 1.0,
  "thinking": {
    "type": "adaptive"
  },
  "max_tokens": 8192
}

{
  "model": "gemini-3-pro-image",
  "size": "1920x1080",
  "quality": "hd",
  "n": 1
}

maxOutputTokens limit

Cherry Studio may send very large maxOutputTokens values (e.g., 128k) that exceed Gemini’s limits:

Antigravity automatically limits maxOutputTokens to 65536 for Claude protocol requests to prevent 400 errors.

This is handled automatically - no configuration needed.

Troubleshooting

Empty responses with web search

Problem: Web search returns no content in Cherry StudioFixed in: v4.1.21Solution:

Upgrade Antigravity to v4.1.21 or later
Restart both Antigravity and Cherry Studio
Test web search again

Technical details: Antigravity now sends complete SSE event lifecycle required by Cherry Studio.

Thinking mode 400 errors

Problem: 400 INVALID_ARGUMENT with thinking modelsFixed in: v4.1.21Solution:

Ensure Antigravity v4.1.21+
Use claude-sonnet-4-6-thinking model name
Set thinking type to “adaptive” in Cherry Studio
Antigravity automatically converts to compatible format

Connection errors

Problem: Cannot connect to AntigravitySolutions:

Verify base URL: http://127.0.0.1:8045/v1
Check Antigravity proxy is running
Test with: curl http://127.0.0.1:8045/health
Check firewall settings

Image generation errors

Problem: Image generation fails or times outSolutions:

Check account has image generation quota
Verify model name is gemini-3-pro-image
Try smaller image sizes first
Check Antigravity logs for detailed errors
Ensure max body size is sufficient (100MB default)

Citations not appearing

Problem: Web search works but no citations shownSolutions:

Verify Antigravity v4.1.21+
Check Cherry Studio supports citation display
Enable “Show sources” in Cherry Studio settings
Review response format in debug logs

Performance tips

Model selection

Fast responses: Use gemini-3-flash
Best quality: Use claude-sonnet-4-6
Complex reasoning: Use thinking models
Images: Use gemini-3-pro-image

Quota management

Monitor dashboard regularly
Enable quota protection
Use multiple accounts
Set appropriate model routing

Response speed

Keep Antigravity local (same machine)
Use streaming for long responses
Flash models for quick tasks
Reduce max_tokens when possible

Reliability

Add multiple accounts for failover
Enable automatic retry
Configure rate limiting
Monitor error rates

Best practices

Use streaming: Enable streaming in Cherry Studio for better UX
Configure max tokens: Set reasonable limits per model
Enable web search selectively: Only when needed to save quota
Monitor quotas: Check Antigravity dashboard regularly
Test configurations: Verify settings work before heavy use
Keep updated: Both Cherry Studio and Antigravity should be latest versions

Example configurations

Basic setup

{
  "provider": "Antigravity",
  "baseURL": "http://127.0.0.1:8045/v1",
  "apiKey": "sk-antigravity",
  "models": [
    {
      "name": "claude-sonnet-4-6",
      "maxTokens": 4096,
      "temperature": 0.7
    },
    {
      "name": "gemini-3-flash",
      "maxTokens": 8192,
      "temperature": 1.0
    }
  ]
}

With image generation

{
  "models": [
    {
      "name": "gemini-3-pro-image",
      "type": "image",
      "size": "1920x1080",
      "quality": "hd"
    }
  ]
}

With thinking models

{
  "models": [
    {
      "name": "claude-sonnet-4-6-thinking",
      "maxTokens": 8192,
      "thinking": {
        "type": "adaptive"
      }
    }
  ]
}

Getting Started

Core Features

Integration Guides

Configuration

Deployment

Troubleshooting

Overview

Prerequisites

Configuration

Supported features

Chat completions

Extended thinking

Web search with citations

Image generation

Advanced configuration

Model-specific settings

maxOutputTokens limit

Troubleshooting

Performance tips

Model selection

Quota management

Response speed

Reliability

Best practices

Example configurations

Basic setup

With image generation

With thinking models

Build docs developers (and LLMs) love

Getting Started

Core Features

Integration Guides

Configuration

Deployment

Troubleshooting

​Overview

​Prerequisites

​Configuration

​Supported features

​Chat completions

​Extended thinking

​Web search with citations

​Image generation

​Multi-modal input

​Advanced configuration

​Model-specific settings

​maxOutputTokens limit

​Troubleshooting

​Performance tips

Model selection

Quota management

Response speed

Reliability

​Best practices

​Example configurations

​Basic setup

​With image generation

​With thinking models

​Related documentation

Build docs developers (and LLMs) love

Overview

Prerequisites

Configuration

Supported features

Chat completions

Extended thinking

Web search with citations

Image generation

Multi-modal input

Advanced configuration

Model-specific settings

maxOutputTokens limit

Troubleshooting

Performance tips

Best practices

Example configurations

Basic setup

With image generation

With thinking models

Related documentation