UniAI MaaS

Overview

Model as a Service

UniAI MaaS aims to implement a model as a service platform integrated multi AI models, allowing developers to focus on developing AI-related applications.

Retrieval Augmented Generation

The platform supports a vector database, and an object storage service which enables users to upload, parse, and manage office files.

Multi-modals & Multi-models

In addition to text generation models, it also offers multi-modal models such as image generation and recognition.

Application

UniAI MaaS supports custom application development for integration with platforms like WeChat Mini Programs.

About

UniAI, powered by the node.js library of the same name uniai, provides seamless access and management of a diverse range of AI models and utilities. To incorporate UniAI into your project, you can install it via npm or Github.

Frontend Example

Front end: https://github.com/uni-openai/lechat-pro

Integrated Models

UniAI integrates multiple AI models, including:

Samples

Discover how UniAI is utilized and experience it firsthand:

LeChat Mini App

To launch the mini app, use WeChat to scan the following QR code.

LeChat Pro

👍 We recommend: https://lechat.cas-ll.cn

Multi-model Chat
Office file upload and parse
Image generation
Image recognition

System Requirements

Ensure you have the following NPM libs installed:

Node.js (version >= 18) - nvm Installation Guide
Docker & Docker-compose
LibreOffice for document conversion (libreoffice-convert)
pdf-to-img (canvas-node) - Canvas NPM Package

Getting Started

Configuration

Create an .env file at the project root path:

touch ./.env

Fill the .env file with the following environment variables:

# Platform default admin token
ADMIN_TOKEN=

# OPENAI GPT
OPENAI_API= # openai proxy
OPENAI_KEY= # openai key

# Google AI studio
GOOGLE_AI_API= # google proxy
GOOGLE_AI_KEY= # google key

# ZHIPU AI
# ZHIPU_AI_API= #zhipu proxy
ZHIPU_AI_KEY= # zhipu key
GLM_API= # local deployed glm6b

# SPARK
FLY_APP_ID= # iflytek app id
FLY_API_KEY= # ilfytek api key
FLY_API_SECRET= # iflytek api secret

# baidu wenxin workshop
# BAIDU_API=http://192.168.41.52:5300
BAIDU_API_KEY=
BAIDU_SECRET_KEY=

# Moonshot
# MOONSHOT_API=http://192.168.41.52:5400
MOONSHOT_KEY=

# Stable Diffusion
STABLE_DIFFUSION_API=

# Midjourney
MJ_API= # https://github.com/novicezk/midjourney-proxy
MJ_IMG_PROXY= # proxy to discord cdn images
MJ_TOKEN= # mj token

# stability ai
STABILITY_KEY=

# PostgreSQL database
DB_DIALECT=postgres
POSTGRES_HOST=localhost
POSTGRES_PORT=5432
POSTGRES_USER=postgres
POSTGRES_PASS=postgres
POSTGRES_DB=uniai

# Redis cache
REDIS_HOST=localhost
REDIS_PORT=6379
REDIS_PASS=redis
REDIS_DB=0

# WeChat
WX_APP_ID= # wechat miniapp id
WX_APP_SECRET= # wechat miniapp secret
WX_MCH_ID=
WX_PAY_PRIVATE=
WX_PAY_CERT=
WX_PAY_KEY=

# MINIO storage
MINIO_END_POINT=localhost
MINIO_ACCESS_KEY=
MINIO_SECRET_KEY=
MINIO_PORT=9000
MINIO_BUCKET=uniai

# gee code test
GEE_TEST_ID=
GEE_TEST_KEY=

# aliyun SMS account
ALI_KEY_ID=
ALI_KEY_SECRET=
ALI_SMS_TEMPLATE=
ALI_SMS_SIGN=

# For Docker start pgvector
POSTGRES_DATA_PATH=./data

# For Docker start Minio
MINIO_DATA_PATH=./data
MINIO_ROOT_USER=root
MINIO_ROOT_PASS=12345678

Installation Steps

Node-gyp Installation

npm -g install node-gyp

LibreOffice Installation

Ubuntu: sudo apt install libreoffice
Mac: brew install libreoffice

Node-Canvas Support

Reference: Canvas NPM Documentation
Install dependencies as per your operating system.

Using Yarn (Recommended over npm)

npm -g install yarn
yarn

Sharp Support

yarn add sharp --ignore-engines

Starting the Database

For databases like PostgresSQL (pgvector), Docker and Docker-compose can be used for setup:

sudo apt install docker.io docker-compose

Docker Commands for Database Services

Start pgvector: yarn docker up pgvector
Start Redis: yarn docker up redis
Start Minio (local OSS): yarn docker up minio

Important Notes

Ensure proper permissions for Docker volumes.
Configure Minio after Docker initialization.
Default Docker settings are available in .env.

Minio Access:

URL: http://localhost:9000
Default Username: root
Default Password: 12345678

Running UniAI

Development Mode

Initializes the database.

yarn dev

Production Mode

Compile TypeScript files and start the application.

yarn tsc
yarn start

⚠️ Avoid compiling TypeScript files in development mode. Use yarn clean before yarn dev if tsc was previously run.

Cleaning Up

yarn clean

Documentation

Access UniAI's APIs through common Web HTTP methods, including SSE. For detailed documentation, visit UniAI API Documentation.

Available Models

UniAI integrates various AI models, focusing on NLP and CV domains. Specific models need to be deployed independently. Download URLs and guides are provided.

NLP Models

OpenAI GPT, GLM/ChatGLM, IFLYTEK/SPARK

CV Models

OpenAI DALL-E, Stable Diffusion, MidJourney

Future Enhancements

UniAI is planning to expand its capabilities across:

Prediction APIs
Training APIs
Prompting APIs
Resource Management APIs

Contributing

Youwei Huang huangyw@iict.ac.cn

Institute of Intelligent Computing Technology, Suzhou, CAS

License

MIT

uni-openai/uniai-maas