DeepSeek has released a lightweight, static web-based chat interface that connects directly to its API, allowing users to run AI-powered conversations without relying on a backend server. The tool, hosted on GitHub as deepseek-ai/chat-demo, is designed for simplicity, privacy, and portability — appealing to developers and businesses wary of vendor lock-in or cloud dependency.
The app runs entirely in the browser. Users can launch it by opening index.html locally or deploying the built static files to any web server. No installation, build process, or backend infrastructure is required. This reduces deployment complexity and operational costs, particularly for small teams or organisations in regions like the UK, UAE, US, and Kazakhstan where infrastructure flexibility is increasingly valued.
How it works and what it stores
The chat interface uses the browser’s localStorage for API keys and settings, and IndexedDB for conversations and original attachments. Data remains isolated by browser profile and site origin, meaning no cross-user leakage. When users upload files, speak, or send messages, only the relevant content is sent to the configured API service — by default, api.deepseek.com. The local storage model gives users full control over their data, and deletion tools in Settings remove all chat history and attachments locally.
History import and export are handled via ZIP archives. Exports include conversations and locally stored images but exclude API keys, settings, and non-image attachments. Importing replaces all existing local data after validation, enabling secure migration across devices or environments. This model supports data sovereignty, a growing priority in regulated sectors and privacy-conscious markets.
Features beyond basic chat
The demo supports advanced capabilities: streaming replies, message regeneration, branching conversations, and deep thinking mode. It handles attachments via upload, paste, or drag-and-drop, and includes voice input and text-to-speech with real-time audio feedback on mobile. The interface renders Markdown, code with syntax highlighting, KaTeX for mathematical expressions, and Mermaid diagrams — making it suitable for technical use cases.
It supports both English and Simplified Chinese, light and dark themes, and mobile-responsive layouts. Behind the scenes, it uses the deepseek-flash model via Anthropic-compatible Messages and Files APIs. Speech features rely on separate ASR and TTS endpoints. The app expects the upstream API to support extensions for search, link reading, and auto-titling — limiting compatibility with non-conforming services.
Deployment and development flexibility
Developers can run the app locally using Node.js 22.12+ and pnpm. After cloning the repository, running pnpm dev launches a development server on http://localhost:8090. The pnpm build command generates static files in the dist/ folder, suitable for deployment on platforms like Netlify, Vercel, or any static host.
The default API origin can be customised via createConfig({ apiOrigin, devServer }) in rspack.config.ts. This allows enterprises to route traffic through internal proxies or regional endpoints, improving latency and compliance. The MIT license permits redistribution, provided the license file is included — a benefit for vendors embedding the interface.
What it means
This release reflects a shift toward client-centric AI tools that prioritise data control and minimal infrastructure. Unlike cloud-hosted chatbots from Microsoft or Google, this demo keeps user data on-device unless explicitly sent for processing. For UK and UAE enterprises managing GDPR or ADGM compliance, or Kazakhstani firms navigating data localisation rules, such control reduces legal and operational risk.
The lack of a backend also cuts recurring costs and dependency on specific cloud providers. However, reliance on DeepSeek’s API — even if accessed directly — still creates a service dependency. Organisations must assess uptime, pricing, and data handling policies of the API provider, even when the frontend is self-hosted.
The open-source nature and permissive license allow auditing and modification, but users must verify third-party dependencies’ licences separately. For founders and IT leads, this tool offers a template for building compliant, low-cost AI interfaces — especially where transparency and data ownership are non-negotiable.