ccproxy-dolphin 🐬

A Flask-based proxy that unlocks AI freedom - use any Claude-compatible application with xAI's Grok or even modify Claude's system prompts. Named in honor of Eric Hartford's uncensored Dolphin model series, this proxy embodies the spirit of AI experimentation and user agency.

🙏 Standing on the Shoulders of Giants

This project was inspired by and builds upon:

ccproxy - The original Claude-compatible proxy implementation
anthropic-proxy - Another great proxy implementation
All the forks and contributors who've pushed these ideas forward

🎭 See the Difference

Classic Mode (Original Claude)

Liberation Mode 🐬 (Custom Prompt Override)

✨ Key Features

1. Multi-Backend Support

Use xAI's Grok models through Claude Code or any Anthropic-compatible client
Use Anthropic's Claude with system prompt modifications
Switch backends with a simple environment variable

2. System Prompt Liberation 🔓

Override Claude's system prompts while preserving functionality
Remove restrictions and safety rails
Inject your own personality and principles
Configuration-based prompt transformations

3. Complete API Translation

Anthropic ↔️ OpenAI format conversion
Full streaming support
Tool/function calling support
Comprehensive request/response logging

🚀 Quick Start

Prerequisites

Python 3.10+
An xAI API key (get one from https://console.x.ai/)
(Optional) An Anthropic API key for Claude backend

Installation

# Clone the repository
git clone <this-repo>
cd ccproxy-dolphin

# Create virtual environment
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Usage Scenarios

Scenario	Backend	Custom Prompt	Result
1. Anthropic Pass-through	`anthropic`	`false`	Original Claude, no modifications
2. Anthropic Custom	`anthropic`	`true`	Claude API with your system prompt
3. Grok Pass-through	`grok`	`false`	Grok with Claude's personality
4. Grok Liberation 🐬	`grok`	`true`	Grok with custom prompt, no restrictions

1️⃣ Anthropic Pass-through (Original Claude, No Modifications)

# Terminal 1: Start proxy in pass-through mode
export BACKEND=anthropic
export ANTHROPIC_API_KEY="your-anthropic-api-key"
export USE_CUSTOM_PROMPT=false  # No prompt modifications
python unified_proxy.py

# Terminal 2: Use Claude as normal
export ANTHROPIC_BASE_URL=http://localhost:8000
export ANTHROPIC_API_KEY=dummy-key
claude
# Result: Original Claude with all default behaviors

2️⃣ Anthropic with Custom Prompts (Claude with Your Personality)

# Terminal 1: Start proxy with custom prompts
export BACKEND=anthropic
export ANTHROPIC_API_KEY="your-anthropic-api-key"
export USE_CUSTOM_PROMPT=true  # Enable prompt override
python unified_proxy.py

# Terminal 2: Use modified Claude
export ANTHROPIC_BASE_URL=http://localhost:8000
export ANTHROPIC_API_KEY=dummy-key
claude
# Result: Claude API with your custom system prompt

3️⃣ Grok Pass-through (Use Grok, Keep Claude's Personality)

# Terminal 1: Start proxy with Grok backend
export BACKEND=grok
export XAI_API_KEY="your-xai-api-key"
export USE_CUSTOM_PROMPT=false  # Keep Claude's system prompt
python unified_proxy.py

# Terminal 2: Use Grok through Claude interface
export ANTHROPIC_BASE_URL=http://localhost:8000
export ANTHROPIC_API_KEY=dummy-key
claude --mcp-config "" --strict-mcp-config  # Recommended: disable MCP for reliability
# Result: Grok model with Claude's original system prompt

4️⃣ Grok with Custom Prompts (Full Liberation Mode 🐬)

# Terminal 1: Start proxy with everything unlocked
export BACKEND=grok
export XAI_API_KEY="your-xai-api-key"
export USE_CUSTOM_PROMPT=true  # Override system prompt
python unified_proxy.py

# Terminal 2: Use liberated AI
export ANTHROPIC_BASE_URL=http://localhost:8000
export ANTHROPIC_API_KEY=dummy-key
claude --mcp-config "" --strict-mcp-config  # Recommended: disable MCP for reliability
# Result: Grok model with your custom personality, no restrictions

🎛️ Configuration Options

Environment Variables

BACKEND - Choose backend: grok (default) or anthropic
XAI_API_KEY - Your xAI API key (required for Grok)
ANTHROPIC_API_KEY - Your Anthropic API key (required for Anthropic backend)
USE_CUSTOM_PROMPT - Enable custom system prompts (true/false)
CUSTOM_PROMPT_FILE - Path to prompt template (default: system_prompt_template_unrestricted.txt)
PROMPT_CONFIG_FILE - Path to config JSON (default: prompt_config.json)
ENABLE_FULL_LOGGING - Enable detailed logging (default: true)
LOG_DIR - Directory for logs (default: logs/requests)

Customizing System Prompts

Edit prompt_config.json to control transformations:

{
  "system_name": "Your AI Name",
  "model_name_override": "Your Model",
  "remove_claude_references": true,
  "remove_anthropic_references": true,
  "remove_defensive_restrictions": true
}

Create your own prompt template with placeholders:

{{ENV_INFO}} - Environment information
{{MODEL_INFO}} - Model details
{{MCP_INSTRUCTIONS}} - MCP server instructions

🛠️ Available Proxies

1. `unified_proxy.py` - The Swiss Army Knife

Supports both Grok and Anthropic backends
Full system prompt customization
Best for experimentation

2. `grok_proxy_openai.py` - OpenAI Format Converter

Grok-only, uses OpenAI format internally
Most reliable for tool calling
Production-ready

3. `grok_proxy.py` - Direct Anthropic Format

Original implementation
Direct Anthropic endpoint usage
Has some tool compatibility issues

📝 Example: Removing All Restrictions

# Start proxy with unrestricted template
export USE_CUSTOM_PROMPT=true
export CUSTOM_PROMPT_FILE=system_prompt_template_unrestricted.txt
python unified_proxy.py

# Claude will now:
# - Have no refusal behaviors
# - Respect user agency
# - Focus on truth-seeking and problem-solving

💰 Cost Warning

This proxy uses your API credits:

xAI/Grok: Monitor usage at https://console.x.ai/
Anthropic: Standard Claude API pricing applies

🔍 Debugging

Check unified_proxy.log for request routing
Check logs/requests/ for detailed request/response data
Use ENABLE_FULL_LOGGING=false to disable detailed logging

🚧 Known Limitations & Caveats

Current Issues with Grok Backend

Performance: Grok can be slow compared to Claude, especially for complex requests
Cost: Currently expensive - monitor your usage carefully at https://console.x.ai/
Tool Calling Reliability: MCP tools seem unreliable, so we recommend:
- Disable MCP servers with --mcp-config "" --strict-mcp-config
- Use only the basic built-in tools
Prompt Optimization Needed: Grok needs different prompting strategies than Claude
- Current system prompts are optimized for Claude
- Grok may need more explicit instructions or different formatting
- The proxy could be enhanced to transform prompts for better Grok compatibility

📢 Call to xAI Team

If someone at xAI is reading this:

I'd love early access to test with a Grok-4 coding fine-tune
Happy to help build "Grok Code" - a dedicated coding assistant
This proxy proves there's demand for Grok in coding workflows
Let's make Grok the best AI for developers! 🚀

Contact: @nkeilar

Workarounds

For best results, use simple, direct prompts
Avoid complex multi-step tasks until tool calling improves
Consider using the Anthropic backend for critical work
Experiment with custom prompts optimized for Grok's style

🤝 Contributing

This is a weekend experiment born from late-night hacking and 90's electronic music. See CONTRIBUTING.md for the full vibe.

📜 License

MIT - See LICENSE

📚 Responsible Use

Disclaimer: This is a personal project created for research and experimentation. It does not represent the views or positions of Anthropic, xAI, or any other companies. Use at your own discretion.

ccproxy-dolphin is designed as a professional developer productivity tool. While it enables system prompt customization, users should:

Respect API Provider Terms: Ensure your usage complies with both Anthropic's and xAI's terms of service
Use for Legitimate Purposes: This tool is intended for improving developer workflows, not bypassing safety measures for harmful purposes
Professional Context: Best suited for technical tasks where concise, direct responses improve productivity
Consider Impact: Remember that AI assistants' default behaviors often exist for good reasons

Our Position

This project explores the balance between:

Developer Productivity: Removing unnecessary verbosity that slows down technical work
User Agency: Giving developers control over their tools
Provider Flexibility: Supporting multiple AI backends for redundancy and choice
Responsible Innovation: Pushing boundaries while respecting ethical considerations

We believe developers should have tools that respect their expertise while maintaining professional standards.

🐬 Why "Dolphin"?

Named in honor of Eric Hartford's Dolphin models - the uncensored LLMs that championed user agency and freedom. This proxy continues that philosophy: your AI, your rules.

Built with ❤️ for developer productivity and responsible innovation

Name		Name	Last commit message	Last commit date
.env.example		.env.example
.gitignore		.gitignore
CLAUDE.md		CLAUDE.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
Task.md		Task.md
classic.png		classic.png
converter.py		converter.py
grok_proxy.py		grok_proxy.py
grok_proxy_openai.py		grok_proxy_openai.py
override.png		override.png
prompt_config.json		prompt_config.json
requirements.txt		requirements.txt
system_prompt_parser.py		system_prompt_parser.py
system_prompt_template.txt		system_prompt_template.txt
system_prompt_template_unrestricted.txt		system_prompt_template_unrestricted.txt
unified_proxy.py		unified_proxy.py

Latest commit

History

Folders and files

Repository files navigation