yy838255939/duozhi-workbench

Duozhi Workbench is worth a look if you already juggle several AI website tabs and want one prompt sent to all of them. It is an early 0.1.0 source release, and the README itself warns that some inherited integrations may no longer match the current websites, so treat it as an experiment rather than a settled daily driver.

The problem

You want a second opinion from several chatbots without copying the same prompt into each tab and manually stitching the answers together. The usual workflow means logging in to each site, pasting the question, switching tabs and tracking which reply came from where. Duozhi Workbench replaces that with a split desktop view of real AI websites, a shared prompt box and a group view that collects selected replies on one page. The result is one place to compare answers while each site still handles its own sign-in and file uploads.

What's different

The core mechanism is deliberate: panels load the actual official websites rather than calling platform APIs. That means you keep each service's own sign-in, file upload and conversation history, and the project does not need API keys. The trade-off is that the app depends on those pages staying usable. The README says the model adapters come from older public source code and that some platforms have changed domains, APIs or sign-in flows, and real account conversations have not been checked for every adapter, so a panel loading is not proof that prompt delivery works.

The group-chat mode adds a shared prompt that can be sent to visible panels, mention support and a bounded sequential relay that visits each selected AI once. Group replies are collected on one page, and members receive the current request plus recent group history. Groups on the same platform share its desktop sign-in session, so creating a new group does not require another login while that session remains valid. The workspace also retains a browser extension with a prompt library, local history and backup functions, and remembers sidebar order and panel layout locally.

The interface is currently in Simplified Chinese, which narrows who can use it comfortably today even though the README is in English.

Who should try it

This fits developers or multilingual users who regularly compare answers from several AI websites, are comfortable reading a Chinese application UI, and are willing to test each integration themselves. It also suits people who prefer keeping platform accounts in a desktop shell over pasting API keys into a third-party client. It is a poor fit if you need a finished English-language product, if your workflow depends on API access or automation, or if you cannot afford to discover mid-task that one panel no longer delivers prompts correctly. The README's extension path is a separate setup route, but it is a browser extension for the same kind of multi-model workbench rather than a substitute desktop app.

Requirements and limitations

The biggest constraint is compatibility. Because the app drives live websites, each supported platform's sign-in flow and page structure matters, and the README explicitly says legacy adapters have not all been checked against real account conversations. Some integrations may need repair before they are useful. The desktop app is an Electron application; building from source requires Node.js 24.15.0 and Corepack, with the supplied build and start commands, so this is currently a source-first project rather than a packaged installer.

The browser extension requires Chrome, Developer mode and loading the built dist folder manually. Keeping dist is necessary because the unpacked extension loads from it, and after source changes you rebuild and reload the extension. Local layout, ordering and history are saved locally, but that does not solve the upstream maintenance problem: when a site changes its sign-in flow or page, the panel is the thing that needs attention. If platform compatibility is central to your use, check the current adapter status against each site you care about before relying on it.

HotRepo's take

The project was created on 2026-10-25 and had 102 total stars when it appeared at #22 on HotRepo's NEW-REPOSITORY board for TypeScript on 2026-10-28. That is a single dated snapshot, not a daily growth measurement, and HotRepo has no comparable daily-board history for it; whether interest persists cannot be judged. Before investing time, verify which model adapters currently work with the official sites you use, since the README itself says that coverage is incomplete.

Project source and documentation