李春川

Posted on Aug 2

Introducing IHUI-AI: 8 Platforms, 176 LLMs, 1 Codebase (Apache 2.0)

#webdev

Introducing IHUI-AI: 8 Platforms, 176 LLMs, 1 Codebase

This is the first article in a 10-part series on building IHUI-AI — an 8-platform open-source AI operating system that unifies 176 large language models.

TL;DR

IHUI-AI is an open-source AI operating system that ships 8 fully-functional client apps (web, API, AI-service, CLI, desktop, browser extension, mobile, mini-program) from a single TypeScript codebase. It unifies 176 large language models (OpenAI, Anthropic, Gemini, DeepSeek, Qwen, GLM, Doubao, Kimi, Ollama, vLLM, ...) behind a 100% OpenAI-compatible API. Apache 2.0 licensed. 5 minutes from git clone to a running instance.

GitHub: https://github.com/IHUI-INF-AI/IHUI-AI

The problem

Building AI products today means juggling 5+ stacks:

Python backend for LLMs (FastAPI + LangChain)
TypeScript frontend (Next.js)
Mobile app (React Native or Swift/Kotlin)
Desktop wrapper (Electron or Tauri)
Browser extension (Chrome MV3)
CLI tool (Node.js or Python)
Mini-program (WeChat / Douyin / Alipay)

Each stack has its own auth, its own state management, its own LLM client. You duplicate the same business logic 5 times, fight 5 different type systems, and ship bugs 5 times faster.

The solution: 1 codebase, 8 clients

IHUI-AI ships 8 client applications from a single TypeScript codebase:

Web (Next.js 15 + React 19 + Tailwind 4)
API (Fastify 5, 1300+ routes)
AI Service (FastAPI + LangGraph, 50+ Agent graphs)
CLI (Node.js TUI REPL)
Desktop (Tauri 2.0, cross-platform, offline-capable)
Browser Extension (WXT + Chrome MV3)
Mobile (React Native + Expo, iOS + Android)
Mini-Program (Taro 4, WeChat + Douyin + Alipay + Baidu)

All 8 clients share 100% of TypeScript types, database schema, and API contracts. Change a route in apps/api, and the change is immediately typed in web/mobile/desktop/extension/miniapp/CLI.

176 LLM providers in one endpoint

from openai import OpenAI

client = OpenAI(
    api_key="ihui-local-dev-key",
    base_url="http://localhost:8802/v1"  # ← drop-in replacement
)

resp = client.chat.completions.create(
    model="auto-router",  # or any of 176 models
    messages=[{"role": "user", "content": "Hello!"}],
)

The model field accepts any of:

OpenAI: gpt-4o, gpt-4-turbo, o1, o3, gpt-4o-mini
Anthropic: claude-opus-4, claude-sonnet-4, claude-haiku-4
Google: gemini-2.0-pro, gemini-2.0-flash
DeepSeek: deepseek-chat, deepseek-coder, deepseek-reasoner
Qwen: qwen-max, qwen-plus, qwen-turbo, qwen-coder
GLM: glm-4-plus, glm-4-flash, glm-4-coder
Doubao: doubao-pro, doubao-lite
Kimi: moonshot-v1-128k, kimi-k2
Ollama (local): llama3.3, qwen2.5, mistral
vLLM (local): any open-weights model
... and 156 more

Production-grade, not a toy

Metric	Value
Database tables	340 (with multi-tenant RLS)
Migrations	144
API endpoints	1300+
LangGraph Agent graphs	50+
MCP servers bundled	100+
GitHub Actions workflows	21
Pre-commit gatekeepers	33+
Test suites / cases	237 / 5346
E2E specs	63
API P99 latency	< 80ms
Throughput	12k req/sec on 4-core

Quickstart

git clone https://github.com/IHUI-INF-AI/IHUI-AI
cd IHUI-AI
pnpm install
pnpm dev  # starts web (8801) + API (8802) + AI service (8803)

Open https://ihui.ai for the live demo.

What's next

In the next 9 articles, I'll deep-dive into:

LangGraph state machine production patterns
MCP gateway implementation
8-platform TypeScript sharing strategies
LiteLLM routing across 176 models
Multi-tenant RLS + audit chain
Tauri desktop engineering
React Native + Expo mobile patterns
Taro 4 mini-program multi-platform
Open-source business models

Follow me on dev.to or subscribe to the IHUI-AI newsletter to not miss them.

License

Apache-2.0 — free for commercial use, no restrictions.

🤖 Generated with love by the IHUI-AI team. Star us on GitHub if this was useful!

DEV Community