What is an AI desktop agent?

An AI desktop agent is an app on your own computer where an AI agent works directly with your files and folders. It can sort documents, read PDFs, research a topic or write code.

The difference from a chat is access. A chatbot sees what you upload, one file at a time, and hands you back text to save somewhere. A desktop agent works where your files already are: it can open a folder, read a hundred PDFs, rename the photos, write the summary to a new file and tell you what it did.

What people use desktop agents for

• Files and folders: rename many files to a pattern, sort downloads by type, compare two folders.
• Documents: read a stack of PDFs and pull out dates, totals or clauses into one table.
• Photos: organise by date, find duplicates, prepare a set for sharing.
• Research: search the web on a question and save notes and sources to a file you keep.
• Code: for developers, the same agent can build, test and fix a project.

Local-first AI agent: what it means, and what it does not

A local-first AI agent is one that runs on your computer and works directly in your own files, terminal and tools, rather than in a hosted web workspace you upload your work into. It helps to separate two layers:

• Where the agent runs: the program that reads your files, runs commands and keeps your history. In a local-first agent, that runs on your computer.
• Where the model runs: the AI model that does the thinking. In Mel, those are cloud models (Claude, GPT, Gemini, DeepSeek and Qwen), so Mel needs an internet connection.

Running on your computer still makes a big difference: your project stays on your disk, in your folders, under your version control. Content a task needs is sent to the model provider you use, under that provider's terms.

What to check before you give an agent your files

1. Control. Can you decide which actions need your approval, especially deleting and moving?
2. Visibility. Does it show what it changed?
3. Which providers receive content. Know which model companies the agent calls.
4. Backups. Before a big reorganisation, copy the folder.
5. Model choice. Different models are better at reading, writing or reasoning.

How Mel does it

• It runs on your computer, Mac, Windows or Linux, in your own folders.
• Not only code. Research a topic, read a pile of PDFs, sort photos, rename files.
• Deletes and moves stay in your hands. Configurable approvals decide which commands and edits run.
• Cloud models do the thinking (Claude, GPT, Gemini, DeepSeek or Qwen), so Mel needs a connection.

How the options compare

A safe first task

Pick a folder you have copied. Ask the agent to list what it would change before it changes anything ("propose new names for these files, do not rename yet"). Check the list, then let it do the job.

FAQ

What is an AI desktop agent? An app on your computer where an AI agent works directly with your files and folders: documents, PDFs, photos, research and code.

What is a local-first AI agent? An agent that runs on your computer and works in your own files and tools, instead of a hosted web workspace. Mel runs on your computer and uses cloud AI models.

Does Mel need an internet connection? Yes. Mel's agents use cloud models, so they need a connection. Your terminal and files are on your computer.

Is an AI desktop agent only for programmers? No. Sorting files, reading PDFs and research need no code. Mel looks like a terminal, so it suits people comfortable seeing a command line.

Can it delete my files? Only if your approval settings allow it. Keep deletes and moves behind an approval, and keep a backup.

• What are AI agent approvals?
• What is an agentic terminal?
• What is a vibe coding app?
• Mel vs Goose
• Mel vs Claude Code