DeepSeek open-sources chat frontend: no server needed, but you still need your own API key
DeepSeek has released chat-demo, a purely static chat frontend. Search, document parsing, speech, and local storage all live in the browser, but calling the model still requires your own API key—and keeping chat history local doesn't mean content isn't uploaded.
DeepSeek has open-sourced chat-demo, with the repository at https://github.com/deepseek-ai/chat-demo. It is a purely static chat frontend: you don't need to set up your own backend, and the browser connects directly to the DeepSeek API. The official online demo is at https://deepseek-ai.github.io/chat-demo/, and a Release archive is also available; after unzipping, double-click index.html to open it locally. On first use, go to settings and enter your own API key.
For people who just want to use it, the “zero-code” claim basically holds true; for people who want to modify it, Node.js 22.12+ and pnpm are the entry requirements. The README lists the development commands: pnpm install, pnpm dev, pnpm build, with build output in dist/. “No need to buy a server” means you don't need to build your own application backend; public access still requires static hosting and a domain name.
The pain point it addresses is very specific. In the past, building a chat interface with search, document reading, and speech required layering frontend, backend, storage, plugins, and deployment one by one. chat-demo packs these into a static site: multi-turn conversations with automatic titles, renaming, deletion, prompt editing, regeneration, and message branch switching; streaming replies, deep thinking, intelligent search, link reading, and source citations; drag-and-drop upload and parsing for PDF, Word, PPT, and images, plus URL body-text extraction; real-time audio input on mobile, with answers that can be read aloud; support for Markdown, code highlighting, KaTeX, and Mermaid; and Chinese/English, light/dark themes, mobile layouts, and history import/export.
Data needs to be looked at separately. The API key and settings are stored in browser localStorage, while conversations and original attachments are stored in IndexedDB, isolated by browser configuration and site origin. Chat history stays local by default; history import/export uses ZIP, and importing replaces all local conversations and attachments. However, when you send messages, upload attachments, or use speech, the relevant content is sent to the configured API service. “Local” refers to where archives are stored, not an offline model. Clearing browser cache may also wipe the archive. Some commenters warn: don't put it on the public internet for others to share.
Technical details are in the README: the default API address is https://api.deepseek.com, the default model is deepseek-flash, and it uses the Anthropic-compatible Messages / Files API; speech uses ASR / TTS endpoints. Search, link reading, titles, and latest_reminder all rely on request extensions, and the server side must support the corresponding capabilities. The license is MIT.
Some commenters say the zero-code part is genuinely appealing, but the domain-name step can take beginners a while. That assessment is fair. The value of chat-demo is more that it turns official capabilities into a decomposable frontend sample: people who want to use it quickly can enter a key and run it; people who want to modify it can directly see how SSE, storage, history import/export, speech, and the Markdown pipeline are wired up. What it saves is the time spent building the scaffolding; you still pay the model usage fees yourself. The open-source spirit isn't about being free—it's about putting the parts on the table.
Original tweet: https://x.com/huangyun_122/status/2121894829596815502
发布时间: 2026-11-16 05:25