NextChat

What NextChat does

NextChat, originally called ChatGPT-Next-Web, is a compact chat front end for hosted language models. You deploy it as a small Next.js site, often with one click on Vercel, or run it in a single Docker container, or install the Tauri desktop app on Windows, macOS or Linux. It has the essentials of a ChatGPT-style assistant: conversation history, image input for vision models, Markdown with LaTeX and Mermaid, prompt templates the project calls “masks”, and automatic compression of long chats to save tokens.

Later versions added plugins (web search, a calculator and arbitrary APIs), MCP tool support, an artifacts window for previewing generated pages, realtime voice chat and image generation through Stability. Document chat is not included; a local knowledge base is still an unchecked item on the roadmap.

Models and providers

The server is configured with API keys through environment variables. Supported providers include OpenAI, Azure OpenAI, Google Gemini, Anthropic, DeepSeek, xAI, Moonshot, SiliconFlow and several Chinese clouds (Baidu, ByteDance, Alibaba, Tencent, iFlytek, ChatGLM). Changing BASE_URL points it at any OpenAI-compatible server, and the README names LocalAI and RWKV-Runner for self-hosted models. There is no built-in local engine.

Being open source does not change where prompts go: with an OpenAI, Anthropic or Google key, conversations are processed by that company. Users can also enter their own keys in the settings unless the operator hides that option.

Privacy and data

According to the README, all chat data stays in the browser; optional sync uses an UpStash Redis database or a WebDAV server you supply. There is no server-side database.

One detail is not in the README. In the web build, the page layout loads Google Analytics whenever a measurement ID is configured, and the server falls back to an ID hard-coded in the source when the GA_ID variable is unset (an empty value falls back too). A self-hosted web deployment therefore loads Google Analytics by default. Set GA_ID to your own property or remove the code if you want to avoid it. Vercel Speed Insights is added only on Vercel. We did not check whether the desktop app behaves the same way.

Self-hosting NextChat

Requirements are minimal: Node.js 18 or Docker 20. The official image is yidadaa/chatgpt-next-web and takes the provider keys, an optional proxy and the CODE variable as environment variables. CODE sets one or more shared access passwords. There are no user accounts, roles or SSO, and the FAQ warns that without CODE anyone who finds the deployment can use your API keys. MCP can only be switched on when CODE is set. Vercel users must fork the repository and enable the sync workflow to receive updates.

Hosted version

The developers run NextChat AI at nextchat.club, where you either bring your own API credentials or buy pay-as-you-go credits. A paid Enterprise Edition for private deployment adds branding, an admin panel, permissions, a knowledge base and auditing; it is not part of the open source code.

Limitations

NextChat is a client, not a platform: no accounts, no document chat, no built-in search engine integration beyond plugins, and local models only through another server. The Google Analytics default is a real privacy issue for self-hosters who do not read the source. Team features sit in the commercial edition.

Who it suits

NextChat suits individuals and small groups who use hosted APIs with their own keys and want a fast, simple interface on the web or desktop without running a database. Teams that need logins and permissions should look at LibreChat; people who want a lightweight browser-only client with more advanced features can compare Big-AGI. The AI chat guide covers the wider field.