OpenDeep [v0.9]

Unofficial Python Client for DeepSeek API

Access DeepSeek's powerful AI models with a clean, Pythonic interface — no official API key required.

Features • Installation • Quick Start • Documentation • Examples

---

Overview

OpenDeep is a lightweight, unofficial Python library that provides programmatic access to DeepSeek's AI models by reverse-engineering the chat.deepseek.com web interface. It offers a simple, intuitive API inspired by Google's Generative AI SDK.

Why OpenDeep?

• No API Key Required — Uses your existing DeepSeek web account token
• Full Feature Parity — Access all models including Pro, Flash, Reasoner, and Expert
• Thinking Mode — Enable DeepThink reasoning for complex problems
• Web Search — Integrated search capabilities (Flash model)
• Streaming Support — Real-time response streaming with colored reasoning output
• Async Ready — Full async/await support for modern applications
• Cloudflare Bypass — Optional curl_cffi integration for reliable access

---

Features

Supported Models

Model	Description	Thinking	Search	Best For
deepseek-v4-pro	Latest flagship model	Default	No	Complex reasoning, coding, analysis
deepseek-v4-flash	Fast & efficient	Optional	Yes	Quick responses, web search tasks
deepseek-chat	General purpose	Optional	No	Everyday conversations
deepseek-reasoner	Reasoning-focused	Default	No	Math, logic, step-by-step problems
deepseek-expert	Expert mode	Optional	No	Specialized tasks

Core Capabilities

• Single-turn Completions — Quick question-answer interactions
• Multi-turn Conversations — Maintain context across multiple messages
• Streaming Output — Watch responses generate in real-time
• Thinking Visualization — See the model's reasoning process (gray text)
• Proof of Work — Automatic challenge solving for API access
• Session Management — Automatic chat session handling

---

Installation

Basic Installation

pip install opendeep

With Cloudflare Bypass (Recommended)

pip install opendeep[cf]

This installs curl_cffi which provides TLS fingerprinting to bypass Cloudflare protection.

From Source

git clone https://github.com/cmpdchtr/opendeep.git
cd opendeep
pip install -e .

Requirements

• Python 3.10+
• curl_cffi>=0.7.0 (recommended)
• requests>=2.31.0
• wasmtime>=18.0.0
• numpy>=1.26.0

---

Quick Start

1. Get Your API Token

1. Visit chat.deepseek.com
2. Log in to your account
3. Open browser DevTools (F12)
4. Go to Application → Local Storage → https://chat.deepseek.com
5. Find the userToken value
6. Copy the token value

2. Configure and Use

import opendeep

# Configure with your token
opendeep.configure(api_key="your_token_here")

# Create a model
from opendeep import GenerativeModel
model = GenerativeModel("deepseek-v4-pro")

# Generate content
response = model.generate_content("Explain quantum computing in simple terms")
print(response.text)

---

Documentation

Configuration

import opendeep

opendeep.configure(
    api_key="your_token",              # Required: userToken from localStorage
    base_url="https://chat.deepseek.com/api/v0",  # Optional: custom endpoint
    user_agent="Custom Agent/1.0"      # Optional: custom user agent
)

Synchronous API

Single-turn Completion

from opendeep import GenerativeModel

model = GenerativeModel("deepseek-v4-pro")
response = model.generate_content("What is the meaning of life?")
print(response.text)

Multi-turn Conversation

from opendeep import GenerativeModel

model = GenerativeModel("deepseek-v4-pro")
chat = model.start_chat(thinking_enabled=True)

# First message
response1 = chat.send_message("What is Python?")
print(response1.text)

# Follow-up (context is maintained)
response2 = chat.send_message("How do I install it?")
print(response2.text)

# Another follow-up
response3 = chat.send_message("Show me a hello world example")
print(response3.text)

Streaming Output

from opendeep import GenerativeModel

model = GenerativeModel("deepseek-v4-pro")

# Stream output to console with colored reasoning
model.generate_content(
    "Write a poem about artificial intelligence",
    stream=True,
    thinking_enabled=True
)

Web Search (Flash Model Only)

from opendeep import GenerativeModel

model = GenerativeModel("deepseek-v4-flash")
response = model.generate_content(
    "What are the latest developments in AI?",
    search_enabled=True
)
print(response.text)

Asynchronous API

Basic Async Usage

import asyncio
from opendeep import AsyncGenerativeModel

async def main():
    model = AsyncGenerativeModel("deepseek-v4-pro")
    response = await model.generate_content("Hello, DeepSeek!")
    print(response.text)

asyncio.run(main())

Async Multi-turn Conversation

import asyncio
from opendeep import AsyncGenerativeModel

async def main():
    model = AsyncGenerativeModel("deepseek-v4-pro")
    chat = model.start_chat(thinking_enabled=True)
    
    response1 = await chat.send_message("What is machine learning?")
    print(response1.text)
    
    response2 = await chat.send_message("Give me an example")
    print(response2.text)

asyncio.run(main())

Concurrent Requests

import asyncio
from opendeep import AsyncGenerativeModel

async def main():
    model = AsyncGenerativeModel("deepseek-v4-pro")
    
    # Run multiple requests concurrently
    tasks = [
        model.generate_content("Question 1"),
        model.generate_content("Question 2"),
        model.generate_content("Question 3"),
    ]
    
    responses = await asyncio.gather(*tasks)
    for i, response in enumerate(responses, 1):
        print(f"Response {i}: {response.text[:100]}...")

asyncio.run(main())

Advanced Options

Thinking Mode

Enable DeepThink reasoning for complex problems:

# Thinking is enabled by default for Pro and Reasoner models
model = GenerativeModel("deepseek-v4-pro")  # thinking_enabled=True by default

# Explicitly enable/disable
response = model.generate_content(
    "Solve this math problem: ...",
    thinking_enabled=True  # or False to disable
)

Model Selection

from opendeep import GenerativeModel

# Pro model (best quality, slower)
pro_model = GenerativeModel("deepseek-v4-pro")

# Flash model (fast, supports search)
flash_model = GenerativeModel("deepseek-v4-flash")

# Reasoner model (optimized for reasoning tasks)
reasoner_model = GenerativeModel("deepseek-reasoner")

# Expert model (specialized)
expert_model = GenerativeModel("deepseek-expert")

---

Examples

Code Generation

from opendeep import GenerativeModel

model = GenerativeModel("deepseek-v4-pro")

prompt = """
Write a Python function that:
1. Takes a list of numbers
2. Returns the moving average with window size 3
3. Includes type hints and docstring
"""

response = model.generate_content(prompt, thinking_enabled=True)
print(response.text)

Data Analysis

from opendeep import GenerativeModel

model = GenerativeModel("deepseek-v4-pro")

data = """
Sales data for Q1 2024:
- January: $45,000
- February: $52,000
- March: $48,000
"""

response = model.generate_content(
    f"Analyze this sales data and provide insights:\n{data}",
    thinking_enabled=True
)
print(response.text)

Web Search Integration

from opendeep import GenerativeModel

model = GenerativeModel("deepseek-v4-flash")

response = model.generate_content(
    "What are the top 5 programming languages in 2024?",
    search_enabled=True
)
print(response.text)

Streaming with Progress

from opendeep import GenerativeModel

model = GenerativeModel("deepseek-v4-pro")

print("Generating response...")
print("-" * 50)

# Stream shows reasoning in gray, then final response
model.generate_content(
    "Explain the theory of relativity",
    stream=True,
    thinking_enabled=True
)

print("-" * 50)
print("Done!")

---

Architecture

How It Works

┌─────────────────────────────────────────────────────────────┐
│                         User Code                           │
│  model.generate_content("Hello")                            │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                      GenerativeModel                        │
│  - Manages HTTP sessions                                    │
│  - Handles authentication                                   │
│  - Coordinates request/response                             │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                        ChatSession                          │
│  - Creates chat session                                     │
│  - Manages conversation context                             │
│  - Handles multi-turn interactions                          │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                      Proof of Work                          │
│  - Requests POW challenge from API                          │
│  - Solves challenge using WebAssembly                       │
│  - Adds solution to request headers                         │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                      DeepSeek API                           │
│  POST /chat/completion                                      │
│  - SSE streaming response                                   │
│  - Thinking + content chunks                                │
└─────────────────────────────────────────────────────────────┘

Proof of Work (POW)

DeepSeek uses a Proof of Work mechanism to prevent abuse. OpenDeep solves these challenges automatically using a WebAssembly module compiled from Rust. The POW solver:

1. Receives a challenge from the API
2. Computes a valid hash using the WASM module
3. Includes the solution in the x-ds-pow-response header

This happens transparently — you don't need to handle it manually.

---

Troubleshooting

Cloudflare Protection

If you encounter Cloudflare blocks:

# Install with curl_cffi
pip install opendeep[cf]

The curl_cffi library provides TLS fingerprinting that mimics Chrome 120, bypassing most Cloudflare protections.

Invalid Token

If you get authentication errors:

1. Re-login to chat.deepseek.com
2. Refresh the page
3. Get a fresh userToken from localStorage
4. Update your configuration

Rate Limiting

DeepSeek may rate limit requests. If you encounter this:

• Add delays between requests
• Reduce concurrent requests
• Wait a few minutes before retrying

Missing Dependencies

If you see import errors:

# Install all dependencies
pip install curl_cffi requests wasmtime numpy

---

Security Considerations

API Token Security

• Never commit tokens to version control
• Use environment variables in production:

import os
import opendeep

token = os.environ.get("DEEPSEEK_TOKEN")
if not token:
    raise ValueError("DEEPSEEK_TOKEN environment variable not set")

opendeep.configure(api_key=token)

Token Rotation

• Tokens may expire or be invalidated
• Re-login to get a fresh token if requests fail
• Consider implementing automatic token refresh

---

Contributing

Contributions are welcome! Please feel free to submit issues and pull requests.

Development Setup

# Clone the repository
git clone https://github.com/cmpdchtr/opendeep.git
cd opendeep

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

# Install in development mode
pip install -e .

# Install dev dependencies
pip install pytest pytest-asyncio ruff pyright

Code Quality

# Format code
ruff format opendeep/

# Lint code
ruff check opendeep/ --select ALL --ignore ANN401,D203,D213 --fix

# Type check
pyright opendeep/

# Run tests
pytest tests/ -v

---

Limitations

• Unofficial API — This library reverse-engineers the web interface; it may break if DeepSeek changes their API
• No Official Support — Use at your own risk
• Rate Limits — Subject to DeepSeek's rate limiting
• Token Expiry — Web tokens may expire and need refresh

---

License

This project is licensed under the MIT License — see the LICENSE file for details.

---

Disclaimer

This is an unofficial library and is not affiliated with, endorsed by, or supported by DeepSeek. Use responsibly and in accordance with DeepSeek's terms of service.

---

made with hate to corps by @cmpdchtr

Report Bug • Request Feature