A 5MB cross-platform client for Claude, GPT-4, and self-hosted LLMs

What it does

NextChat is a chat frontend that wrangles a wide roster of LLM APIs into one interface. It runs as a web app, PWA, iOS app, or compact desktop client built with Tauri, and keeps conversation data local rather than in the cloud. You can point it at OpenAI, Anthropic, Google, Azure, Baidu, ByteDance, Alibaba, iFlytek, or your own LocalAI and RWKV-Runner instance.

The interesting bit

The app treats providers as interchangeable endpoints rather than walled gardens. It compresses long chat histories automatically to save tokens, lets you share conversations as images or to ShareGPT, and recently added plugin and MCP support for extending capabilities beyond simple text generation.

Key highlights

• Broad backend support: commercial APIs from OpenAI, Claude, Gemini, DeepSeek, Azure, Baidu, ByteDance, Alibaba, and iFlytek, plus self-hosted models.
• Cross-platform surface: web, PWA, iOS, and ~5MB desktop clients for Linux, Windows, and macOS via Tauri.
• Local-first privacy: data lives in the browser or local storage; the app is essentially just a UI layer over your own keys.
• Conversation utilities: automatic history compression, prompt templates (“masks”), artifact previews, and ShareGPT export.
• Enterprise tier adds private deployment, brand customization, admin controls, and security auditing for internal knowledge bases.

Caveats

• MCP support is experimental and gated behind the ENABLE_MCP=true build flag.
• The iOS source repository is advertised as “coming soon” and not yet available.
• One-click Vercel deploys create a standalone project rather than a fork, so automatic upstream updates require manual workflow configuration.

Verdict

Ideal for developers and teams who already have API keys scattered across providers or who self-host models. If you need a fully managed AI service with built-in models, this is just a polished aggregator—not a replacement.