Can AI edit your files without uploading them?

The usual way to get AI help with a file is to attach it to a chat. That works, but be clear about what happens: the file is transmitted to the AI provider's servers, processed there, and retained for however long its policy says. For a holiday photo that may be fine. For contracts, financial spreadsheets, medical letters or unreleased design work, it often is not — and in many workplaces it is explicitly against policy.

The alternative is to flip the direction: instead of sending your files to the AI, let the AI operate software that runs where your files already are — on your computer. EditItAll's editors already process everything on your device; the missing piece was a way for an AI assistant to drive them. That piece is a small local connector.

The connector uses the Model Context Protocol (MCP), an open standard that lets AI assistants call tools on your machine. Once installed, your assistant gains a set of EditItAll tools: write cells and formulas into the spreadsheet editor and read back computed results, load or export whole CSVs, open a PDF and extract its text page by page, rotate or merge pages, draft an illustration into the vector editor, build a slide deck from an outline, and push images through the same compressors the image tools use.

Everything runs locally: the connector opens the editors in a private browser session on your computer and operates them there. Your files are never uploaded to EditItAll. One honest nuance — whatever a tool returns (a cell value you asked for, a page of extracted text, a screenshot you requested) does go to the AI assistant you are chatting with, because that is the answer you asked it for. You stay in control of what that is, and the files themselves never leave your machine.

Setup takes about a minute: download the extension from the AI connector page, drag it into Claude Desktop's Settings → Extensions, and make sure Google Chrome or Microsoft Edge is installed. Then just ask, in plain language:

• “Compress every PNG in my screenshots folder to WebP at quality 80.”
• “Load sales.csv into the spreadsheet, add a totals row, and export it back.”
• “Open report.pdf, summarize page 3, then delete the last page and save it.”
• “Make a five-slide deck introducing the new project.”
• “Draw a simple rocket illustration and put it in the vector editor.”

The assistant can also take a screenshot of an editor to check its own work, and you can take over in the editor at any point — it is the same photo editor, spreadsheet and PDF editor you would use by hand.

Honesty section: a local connector is not the answer to everything. If you want AI to look at a photo — describe it, generate a variation, remove an object with generative fill — the model genuinely needs the pixels, which means sending the image to it. The same goes for asking an AI to deeply analyze a long document in one shot: pasting or attaching it is simpler than reading it page by page through tools.

The rule of thumb: when the task is operating on files (convert, compress, reorganize, calculate, assemble), a local connector does it without an upload. When the task is understanding content at depth, decide deliberately whether that content is something you are comfortable sending to an AI service — and if not, keep it local and do that part yourself.