Agent FAQ

Short answers to the questions people ask most about Mel's AI agent. Each answer links to the page with the details.

Getting started

What is Mel's agent? An AI coding agent inside the terminal. You describe the task, and it reads your code, edits files, runs commands and checks its own work in the pane you're in. Start one with Ctrl+Enter (⌘Enter on a Mac). See Agent overview.

Do I need an account? For the agent, yes: a free Mel account, created from the sign-in panel in the app. The terminal itself needs no account. See Account.

Is the agent free? On the Free plan the agent runs on your own provider key, at 0 Mel credits; you pay your provider, and Groq and Google AI Studio give free keys. Paid plans include monthly credits so you don't need a key. See Get a free model key and Models & pricing.

Does the agent work offline? No. The agent's loop runs on Mel's server and talks to the model from there, so it needs an internet connection. The terminal, editor and code review work without one.

Models

Which model does the agent use? By default, auto: Mel picks a model for each turn from the turn itself, with no extra model call first. Choose a mode (Balanced, Cost-efficient, Genius, Open-weights) or pin a model from the model chip or /model. See Models & routing.

Can I use my own API key? Yes, on any plan, in Settings ▸ AI ▸ Provider API keys. A turn on your key costs 0 Mel credits. The key stays in your operating system's keychain (a file only you can read on Linux) and is sent only with your own requests.

Why did a different model answer? If the chosen model is unavailable, Mel moves to another one and tells you which model answered.

Safety and control

Can the agent change my files without asking? Not by default. Reads run freely; every edit and every command that changes something shows an approval card first, with the exact diff for an edit. You can tick Always allow for one kind of action in a conversation, or turn on autonomous mode. See Tools & approvals.

What does the agent always ask about? A command that can't be undone (a recursive delete outside the project, git push --force, sudo and more), reading a file that holds credentials, writing outside the project, and any connector call. These wait for you even hands-off. See Security & privacy.

How do I stop the agent? Press the stop button that replaces send while a turn runs. The turn ends at once. Background jobs keep running, and each has its own Stop above the input. See Background commands.

How do I undo what the agent did? Open the code review panel (Ctrl+Shift+=, ⌘⇧= on a Mac) to see every change against git, and discard what you don't want. See Reviewing the agent's changes.

Your code and data

Where do the agent's tools run? On your computer, in the pane's folder (or on the remote host in a melted SSH session). Web search, web pages and connector calls run on Mel's server.

Does my code go to the model? What you and the agent put into the conversation does: the files it reads, command output, your messages. Secret-looking values are masked before a model sees them. See Security & privacy.

Where are conversations saved? As files on your computer, under ~/.local/share/mel/accounts/<account>/conversations/, and synced to your account unless you turn off Settings ▸ Privacy ▸ Sync conversations to my account. See Conversation management.

Does the agent remember things between conversations? Yes, with project memory: short notes per project, which you can see, edit and delete. Turn it off per conversation with the ◈ chip.

Working with the agent

How do I give the agent rules for my project? Put them in a MEL.md file in the repository. Mel also reads AGENTS.md and CLAUDE.md. See Rules and MEL.md.

Can I run several agents at once? Yes. Each tab or pane can hold its own conversation, the Runs screen (Ctrl+Shift+M, ⌃⇧M on a Mac) lists them all, and orchestration lets one lead agent split a task across workers. Give each task its own git worktree so agents don't edit the same files.

How do I know when it's done? The notification bell, a flash card and a desktop notification tell you when an agent finishes or needs you. See Notifications.

Can I use Claude Code or Codex in Mel? Yes. They run in a Mel tab with full keyboard support, and they don't use Mel credits. See Embedded CLI agents.

What happens when a conversation gets too long? Mel compacts it at the next turn boundary, folding older turns into a summary. Type /compact to do it yourself. See Slash commands.

Can I use the agent in VS Code? Yes, with the Mel extension. It uses the Mel app on your computer and your Mel plan.

Still stuck? Write to hello@openmel.dev or use the contact form.