When an AI chat runs in a browser, it usually cannot see the folder you are working in. You can paste in a file, but that gets awkward when the task involves several files, a changing project, or edits you need to review. DeskWink is built to bridge that gap: it connects supported web AI apps to files and other tools on a Windows PC or Mac.
The important distinction is that DeskWink does not provide the AI model. It acts as a connector between an AI app and your computer. The model you use still comes from that app, and whether you can connect DeskWink depends on the provider, plan and region.
A connector between chat and computer
DeskWink combines a small command-line program on your computer with a remote MCP connector address that you add to an AI app. The setup starts in a terminal: install the CLI with npm, launch it, and confirm the pairing in a browser. Then add DeskWink as a custom connector in a compatible app and sign in.
That sequence matters. DeskWink is not just a file picker inside a chat window. The computer running the CLI is what gives the connector access to local resources. The landing page describes file browsing, reading, searching and editing, as well as running commands in local projects. It also describes optional browser control, and desktop control on Windows. These are capabilities of the wider product, not all just file access.
The practical use is straightforward: ask an AI to inspect project files, summarize notes or propose an edit without copying each file into the conversation. The AI sends the request through DeskWink, which acts on the connected computer. You can also send requests from a phone if the computer is on and DeskWink is running. This depends on the AI app showing the connector on mobile.
The DeskWink site says the CLI requires Node.js 22 or newer, Windows 10 or 11, or macOS 13 or later. That makes setup more involved than adding a browser extension alone. It also means DeskWink is a poor fit for someone who wants a purely web-based tool, or who cannot install and run software on the computer holding the files.
Approval is a setting, not a guarantee
There is a consequential difference between “you can approve changes” and “every change waits for approval.” The product description says every file change waits for approval. The landing page, however, says a new installation has everything enabled and nothing waiting for approval. It presents approval levels as configurable, including settings that make changes wait for review.
That means a user should not assume a fresh install is in review-first mode. The landing page says the default scope can include the whole computer, commands, the browser and, on Windows, desktop control. You can limit access to selected folders, disable capabilities or change approval settings, but those controls need attention before relying on them. In particular, commands run with the permissions of your user account and can reach whatever that account can reach.
The product describes file safeguards too. Its file tools cannot open certain protected locations, including SSH keys, credential files and password stores. It says file changes are backed up locally and shown in Activity, with a restore command available. It also says file contents pass through DeskWink to the AI app but are not stored in DeskWink’s database or logs. Those are useful design claims, but they do not remove the need to understand what the AI app itself receives or what permissions you have enabled.
Useful when copying files is the bottleneck
DeskWink’s appeal is not that it makes a model smarter. It gives a compatible model a way to work against local material instead of relying on whatever the user pastes into chat. For a developer who wants an AI to inspect a project or run tests in the existing environment, that could reduce repetitive handoffs. For someone working with notes or documents, it offers a route to ask about a folder without uploading each file manually.
The trade-off is broader access and more setup responsibility. You need a supported AI app with custom connectors, a suitable computer, Node.js, and a CLI that stays running. Remote requests also depend on that computer being on. DeskWink is in closed beta and free during the beta, but that does not change the operational question: how much access should a remote AI request be allowed to have?
DeskWink is for people who already use compatible AI chats and want them to work with files or projects on their own Windows PC or Mac. It is not for someone who expects every action to require approval automatically, or who wants to avoid configuring a local program and its permissions. Set the limits first, then decide whether the convenience is worth the access.
Let AI apps work with folders on your computer (closed beta)