New API

🍥 Next-Generation Large Model Gateway and AI Asset Management System

📝 Project Description

[!NOTE]
This is an open-source project developed based on One API

[!IMPORTANT]

This project is for personal learning purposes only, with no guarantee of stability or technical support.

Users must comply with OpenAI's Terms of Use and applicable laws and regulations, and must not use it for illegal purposes.

According to the 《Interim Measures for the Management of Generative Artificial Intelligence Services》, please do not provide any unregistered generative AI services to the public in China.

🤝 Trusted Partners

No particular order

📚 Documentation

For detailed documentation, please visit our official Wiki: https://docs.newapi.pro/

You can also access the AI-generated DeepWiki:

✨ Key Features

New API offers a wide range of features, please refer to Features Introduction for details:

🎨 Brand new UI interface
🌍 Multi-language support
💰 Online recharge functionality (YiPay)
🔍 Support for querying usage quotas with keys (works with neko-api-key-tool)
🔄 Compatible with the original One API database
💵 Support for pay-per-use model pricing
⚖️ Support for weighted random channel selection
📈 Data dashboard (console)
🔒 Token grouping and model restrictions
🤖 Support for more authorization login methods (LinuxDO, Telegram, OIDC)
🔄 Support for Rerank models (Cohere and Jina), API Documentation
⚡ Support for OpenAI Realtime API (including Azure channels), API Documentation
⚡ Support for Claude Messages format, API Documentation
Support for entering chat interface via /chat2link route
🧠 Support for setting reasoning effort through model name suffixes:
1. OpenAI o-series models
  - Add -high suffix for high reasoning effort (e.g.: o3-mini-high)
  - Add -medium suffix for medium reasoning effort (e.g.: o3-mini-medium)
  - Add -low suffix for low reasoning effort (e.g.: o3-mini-low)
2. Claude thinking models
  - Add -thinking suffix to enable thinking mode (e.g.: claude-3-7-sonnet-20250219-thinking)
🔄 Thinking-to-content functionality
🔄 Model rate limiting for users
💰 Cache billing support, which allows billing at a set ratio when cache is hit:
1. Set the Prompt Cache Ratio option in System Settings-Operation Settings
2. Set Prompt Cache Ratio in the channel, range 0-1, e.g., setting to 0.5 means billing at 50% when cache is hit
3. Supported channels:
  - OpenAI
  - Azure
  - DeepSeek
  - Claude

Model Support

This version supports multiple models, please refer to API Documentation-Relay Interface for details:

Third-party models gpts (gpt-4-gizmo-*)
Third-party channel Midjourney-Proxy(Plus) interface, API Documentation
Third-party channel Suno API interface, API Documentation
Custom channels, supporting full call address input
Rerank models (Cohere and Jina), API Documentation
Claude Messages format, API Documentation
Dify, currently only supports chatflow

Environment Variable Configuration

For detailed configuration instructions, please refer to Installation Guide-Environment Variables Configuration:

GENERATE_DEFAULT_TOKEN: Whether to generate initial tokens for newly registered users, default is false
STREAMING_TIMEOUT: Streaming response timeout, default is 300 seconds
DIFY_DEBUG: Whether to output workflow and node information for Dify channels, default is true
FORCE_STREAM_OPTION: Whether to override client stream_options parameter, default is true
GET_MEDIA_TOKEN: Whether to count image tokens, default is true
GET_MEDIA_TOKEN_NOT_STREAM: Whether to count image tokens in non-streaming cases, default is true
UPDATE_TASK: Whether to update asynchronous tasks (Midjourney, Suno), default is true
COHERE_SAFETY_SETTING: Cohere model safety settings, options are NONE, CONTEXTUAL, STRICT, default is NONE
GEMINI_VISION_MAX_IMAGE_NUM: Maximum number of images for Gemini models, default is 16
MAX_FILE_DOWNLOAD_MB: Maximum file download size in MB, default is 20
CRYPTO_SECRET: Encryption key used for encrypting database content
AZURE_DEFAULT_API_VERSION: Azure channel default API version, default is 2025-04-01-preview
NOTIFICATION_LIMIT_DURATION_MINUTE: Notification limit duration, default is 10 minutes
NOTIFY_LIMIT_COUNT: Maximum number of user notifications within the specified duration, default is 2
ERROR_LOG_ENABLED=true: Whether to record and display error logs, default is false

Deployment

For detailed deployment guides, please refer to Installation Guide-Deployment Methods:

[!TIP] Latest Docker image: calciumion/new-api:latest

Multi-machine Deployment Considerations

Environment variable SESSION_SECRET must be set, otherwise login status will be inconsistent across multiple machines
If sharing Redis, CRYPTO_SECRET must be set, otherwise Redis content cannot be accessed across multiple machines

Deployment Requirements

Local database (default): SQLite (Docker deployment must mount the /data directory)
Remote database: MySQL version >= 5.7.8, PgSQL version >= 9.6

Deployment Methods

Using BaoTa Panel Docker Feature

Install BaoTa Panel (version 9.2.0 or above), find New-API in the application store and install it. Tutorial with images

Using Docker Compose (Recommended)


# Download the project
git clone https://github.com/Calcium-Ion/new-api.git
cd new-api
# Edit docker-compose.yml as needed
# Start
docker-compose up -d

Using Docker Image Directly


# Using SQLite
docker run --name new-api -d --restart always -p 3000:3000 -e TZ=Asia/Shanghai -v /home/ubuntu/data/new-api:/data calciumion/new-api:latest

# Using MySQL
docker run --name new-api -d --restart always -p 3000:3000 -e SQL_DSN="root:123456@tcp(localhost:3306)/oneapi" -e TZ=Asia/Shanghai -v /home/ubuntu/data/new-api:/data calciumion/new-api:latest

Channel Retry and Cache

Channel retry functionality has been implemented, you can set the number of retries in Settings->Operation Settings->General Settings. It is recommended to enable caching.