[ 中文 | English ]
Recently, we have found that some self-media guide users to deploy the source code or image of this repository to non-personal use channels and provide services publicly. This behavior may violate DeepSeek's Terms of Use. We hereby remind the relevant self-media and individuals to immediately stop such improper behavior. If the violation continues, DeepSeek reserves the right to pursue legal responsibility.
Supports high-speed streaming output, multi-turn conversation, internet search, R1 deep thinking, and silent deep thinking. Zero-configuration deployment and multi-token support.
Fully compatible with ChatGPT interface.
Here are ten more free APIs for your attention:
Moonshot AI（Kimi.ai）API kimi-free-api
GLM AI API glm-free-api
StepChat API step-free-api
Qwen API qwen-free-api
Metaso AI API metaso-free-api
ByteDance（doubao）API doubao-free-api
ByteDance（jimeng AI）API jimeng-free-api
Spark API spark-free-api
MiniMax（hailuo AI）API hailuo-free-api
Emohaa API emohaa-free-api
- Disclaimer
- Effect Examples
- Preparation
- Docker Deployment
- Render Deployment
- Vercel Deployment
- Native Deployment
- Recommended Clients
- API List
- Notes
- Star History
Reverse-engineered APIs are unstable. It is recommended to use the official DeepSeek API at https://platform.deepseek.com/ to avoid the risk of being banned.
This organization and individuals do not accept any donations or transactions. This project is purely for research and learning purposes!
For personal use only. Do not provide services or commercial use to avoid putting pressure on the official service. Use at your own risk!
For personal use only. Do not provide services or commercial use to avoid putting pressure on the official service. Use at your own risk!
For personal use only. Do not provide services or commercial use to avoid putting pressure on the official service. Use at your own risk!
Please ensure you are in mainland China or have personal computing equipment in mainland China, otherwise, the deployment may not work due to the inability to access DeepSeek.
Get the userToken value from DeepSeek
Start a conversation on DeepSeek, then open the developer tools with F12, and find the value of userToken in Application > LocalStorage. This will be used as the Bearer Token value for Authorization: Authorization: Bearer TOKEN
Currently, the same account can only have one output at a time. You can provide multiple userToken values and concatenate them with ,:
Authorization: Bearer TOKEN1,TOKEN2,TOKEN3
The service will select one for each request.
| Environment Variable | Required | Description |
|---|---|---|
| DEEP_SEEK_CHAT_AUTHORIZATION | No | If configured, the token will be used. If not configured, the Authorization header must be provided in the request |
Pull the image and start the service.
docker run -it -d --init --name deepseek-free-api -p 8000:8000 -e TZ=Asia/Shanghai vinlic/deepseek-free-api:latest
# Or configure the token in the environment variable
docker run -it -d --init --name deepseek-free-api -p 8000:8000 -e TZ=Asia/Shanghai -e DEEP_SEEK_CHAT_AUTHORIZATION=xxx vinlic/deepseek-free-api:latestView real-time service logs
docker logs -f deepseek-free-apiRestart the service
docker restart deepseek-free-apiStop the service
docker stop deepseek-free-apiversion: '3'
services:
deepseek-free-api:
container_name: deepseek-free-api
image: vinlic/deepseek-free-api:latest
restart: always
ports:
- "8000:8000"
environment:
- TZ=Asia/ShanghaiNote: Some deployment regions may not be able to connect to DeepSeek. If the container logs show request timeout or connection failure, please switch to another region! Note: Free account container instances will automatically stop running after a period of inactivity, which may cause a delay of 50 seconds or more for the next request. It is recommended to check Render Container Keep-alive
-
Fork this project to your GitHub account.
-
Visit Render and log in with your GitHub account.
-
Build your Web Service (New+ -> Build and deploy from a Git repository -> Connect your forked project -> Select deployment region -> Select instance type as Free -> Create Web Service).
-
After the build is complete, copy the assigned domain name and append the URL to access.
Note: The request response timeout for Vercel free accounts is 10 seconds, but the interface response is usually longer, which may result in a 504 timeout error from Vercel!
Please ensure that the Node.js environment is installed.
npm i -g vercel --registry http://registry.npmmirror.com
vercel login
git clone https://github.com/LLM-Red-Team/deepseek-free-api
cd deepseek-free-api
vercel --prodPlease install the Node.js environment and configure the environment variables to ensure the node command is available.
Install dependencies
npm iInstall PM2 for process management
npm i -g pm2Build the project. The dist directory indicates the build is complete.
npm run buildStart the service
pm2 start dist/index.js --name "deepseek-free-api"View real-time service logs
pm2 logs deepseek-free-apiRestart the service
pm2 reload deepseek-free-apiStop the service
pm2 stop deepseek-free-apiUse the following secondary development clients to access the free API series projects faster and easier, supporting document/image upload!
LobeChat developed by Clivia https://github.com/Yanyutin753/lobe-chat
ChatGPT Web developed by 时光@ https://github.com/SuYxh/chatgpt-web-sea
Currently supports the /v1/chat/completions interface compatible with openai. You can use openai or other compatible clients to access the interface, or use online services like dify.
Chat completion interface, compatible with openai's chat-completions-api.
POST /v1/chat/completions
The header needs to set the Authorization header:
Authorization: Bearer [userToken value]
Request data:
{
// model name
// default: deepseek
// deep thinking: deepseek-think or deepseek-r1
// internet search: deepseek-search
// deep thinking + internet search: deepseek-r1-search or deepseek-think-search
// silent mode (no output of thinking process or internet search results): deepseek-think-silent or deepseek-r1-silent or deepseek-search-silent
// deep thinking but the thinking process is wrapped with <details> collapsible tags (requires page support for display): deepseek-think-fold or deepseek-r1-fold
"model": "deepseek",
// default multi-turn conversation is implemented based on message merging, which may lead to reduced capabilities in some scenarios and is limited by the maximum token number per round
// if you want to get the native multi-turn conversation experience, you can pass in the id obtained from the previous round of messages to continue the context
// "conversation_id": "50207e56-747e-4800-9068-c6fd618374ee@2",
"messages": [
{
"role": "user",
"content": "Who are you?"
}
],
// if using streaming response, set to true, default is false
"stream": false
}Response data:
{
"id": "50207e56-747e-4800-9068-c6fd618374ee@2",
"model": "deepseek",
"object": "chat.completion",
"choices": [
{
"index": 0,
"message": {
"role": "assistant",
"content": "Greetings! I'm DeepSeek-V3, an artificial intelligence assistant created by DeepSeek. I'm at your service and would be delighted to assist you with any inquiries or tasks you may have."
},
"finish_reason": "stop"
}
],
"usage": {
"prompt_tokens": 1,
"completion_tokens": 1,
"total_tokens": 2
},
"created": 1715061432
}Check if the userToken is alive. If alive, live is true; otherwise, it is false. Please do not call this interface frequently (less than 10 minutes).
POST /token/check
Request data:
{
"token": "eyJhbGciOiJIUzUxMiIsInR5cCI6IkpXVCJ9..."
}Response data:
{
"live": true
}If you are using Nginx to reverse proxy deepseek-free-api, please add the following configuration items to optimize the streaming output effect and improve the experience.
# Disable proxy buffering. When set to off, Nginx will immediately send the client request to the backend server and immediately send the response received from the backend server back to the client.
proxy_buffering off;
# Enable chunked transfer encoding. Chunked transfer encoding allows the server to send data in chunks for dynamically generated content without knowing the size of the content in advance.
chunked_transfer_encoding on;
# Enable TCP_NOPUSH, which tells Nginx to send data as much as possible before sending the data packet to the client. This is usually used in conjunction with sendfile to improve network efficiency.
tcp_nopush on;
# Enable TCP_NODELAY, which tells Nginx not to delay sending data and to send small data packets immediately. In some cases, this can reduce network latency.
tcp_nodelay on;
# Set the keep-alive timeout, here set to 120 seconds. If there is no further communication between the client and the server during this time, the connection will be closed.
keepalive_timeout 120;Since the inference side is not in deepseek-free-api, tokens cannot be counted and will be returned as a fixed number.