DeepSeek chat, but it remembers you and exports ZIPs

What it does Better DeepSeek is a browser extension that patches the official DeepSeek web chat with a hidden system prompt and a set of custom XML-style tags. When the model emits tags like <BDS:create_file> or <BDS:pptx>, the extension intercepts them to generate downloadable files, presentations, spreadsheets, and Word documents directly in the browser. It also adds persistent memory, toggleable skills, roleplay characters, and a LONG_WORK mode that bundles multi-file projects into ZIP archives. The project is unofficial and unaffiliated with DeepSeek AI.

The interesting bit The extension essentially treats the chat UI as a runtime environment: it teaches the model to request web fetches, GitHub repo imports, and even remote MCP tool invocations, then executes those requests locally and injects the results back into context. An experimental “DeepCode” mode attempts to let the model inspect connected workspaces with auto-generated directory and file-read tags, though the README is vague on exactly how those workspaces are mounted.

Key highlights

• Injects a custom system prompt and interprets model-generated tags (<BDS:create_file>, <BDS:excel>, <BDS:chart>, etc.) to produce real downloads.
• Bundles multi-file projects into a single ZIP via the <BDS:LONG_WORK> tag, suppressing chat clutter with a “Working…” indicator.
• Stores persistent key-value memories and markdown “skills” that shape behavior across sessions.
• Supports voice dictation and text-to-speech, plus MCP server integration for external tools over HTTP.
• Adds native navigation fixes, code-block download buttons, and advanced upload options for folders, web pages, and GitHub repos.

Caveats

• The Firefox build is explicitly marked as experimental.
• MCP support is limited to HTTP/Streamable HTTP transport; stdio-based servers are not supported.
• DeepCode and workspace file inspection are experimental, and the README does not clearly explain how local directories are surfaced to the model.

Verdict Worth installing if you do most of your DeepSeek work in the browser and want file generation, memory, and project scaffolding without leaving the tab. If you already pipe DeepSeek through a local API client or an IDE with built-in tooling, this will feel redundant.