What it does

NextChat is a web-first chat interface that talks to pretty much every major LLM API—OpenAI, Anthropic, Google Gemini, DeepSeek, Azure, plus self-hosted models via LocalAI or Ollama. Deploy it free on Vercel in a minute, run the ~5MB desktop build on Linux/Windows/macOS, or install the iOS app. All chat data stays in your browser unless you say otherwise.

The interesting bit

The project treats model providers as interchangeable utilities, not religions. It auto-compresses long conversation history to save tokens, supports plugins (network search, calculator, custom APIs), and recently added MCP support—though you have to flip an ENABLE_MCP=true env var before building, which feels slightly secret-handshake.

Key highlights

• One-click Vercel deploy with password protection via CODE env variable
• Desktop clients built with Tauri, fetching LLM APIs through the native layer for “MORE SECURITY” (their caps)
• Realtime chat support as of v2.15.8, plus Artifacts preview window for generated content
• 15-language i18n including the full CJK set and several European languages
• Plugin ecosystem at NextChat-Awesome-Plugins

Caveats

• The “local knowledge base” feature is on the roadmap but unchecked; RAG isn’t here yet
• Enterprise edition exists as a separate commercial offering with admin panels and audit trails—source repo is the consumer version
• iOS source code is “coming soon” per the README, so the App Store build is currently a black box

Verdict

Grab this if you want a single UI for juggling multiple API keys and self-hosted models without vendor Stockholm syndrome. Skip it if you need deep RAG integration or want a batteries-included managed service.