Quickstart

Five minutes, one loop: run commands, hand a task to the agent, review what changed. This walkthrough assumes you've finished installation.

1. Run some commands

Open Mel. A new tab starts in the agent conversation, so press Esc (or click the "esc return to the terminal" row) to drop to the terminal, which shows the "New terminal session" landing with a few keyboard hints. To start new tabs in the shell instead, turn off Settings ▸ Features ▸ Open new sessions in the Mel agent. Type into the prompt at the bottom and press enter:

cd ~/your-project
ls
git status

Each command becomes a block — the command line, its output, and a header with the directory and how long it took. Notice a few things:

• cd really moves you: the header, the directory chip above the prompt, and every later command follow your new working directory.
• File and directory names in output become links on hover — ctrl+click (cmd+click on macOS) opens them.
• Click a block to select it; its header grows action icons and a right-click menu (copy command, copy output, filter, bookmark).
• As you type, ghost text suggests completions from your session, shell history, and the filesystem. Press the right arrow to accept.

The pills above the input are chips: detected runtime versions (Node, Python, Rust, Go), your current directory, and your git branch with a live change count. They're clickable — the branch chip opens a branch picker, the Node chip a version picker.

2. Start an agent conversation

Type a real question — say, what does the build script in this repo actually do? — and press ctrl+enter (cmd+enter on macOS) instead of enter. The pane becomes an agent conversation and your text is the first message.

The agent streams its answer in place. To figure things out it calls tools — you'll see cards like "Read file", "List directory", "Grep" appear in the transcript as it explores your project. Reading your project, and commands that only read (ls, cat, git status, git diff and the like), run without stopping you.

3. Ask for a change and approve it

Now ask for something that edits code: add a --version flag to the CLI. When the agent wants to write a file or run a command that changes something, it stops and shows an approval card. An edit card shows the diff of exactly what will be written.

• Allow edit (or Run command) — do this one action.
• Always allow … in this conversation — tick the box first, and Mel stops asking for that kind of action for the rest of the conversation.
• Deny (or Esc) — skip it; the agent adapts.

A few steps always need your click, whatever you ticked: a command Mel recognises as impossible to undo, reading a credential file (or running a command that names one), writing outside the project, and any connector call. See Security & privacy.

Every applied edit renders a diff card in the transcript — the file path with +added −removed counts and the changed lines in red/green.

Press esc to drop back to the terminal. The conversation collapses into a small card in your scrollback — click it, or press Ctrl+Shift+Y (⌃⇧Y on a Mac), to continue where you left off.

4. Open code review

Press Ctrl+Shift+= (⌘⇧= on a Mac) (or click Code review in the titlebar). The panel opens beside your terminal showing every uncommitted change as collapsible per-file diff cards — including what the agent just wrote. From here you can open a file in the built-in editor (✎), discard a file's changes, or switch the diff base to compare against another branch.

Happy? Type a message in the panel's commit box and click Commit, or commit from the prompt like you always do: git add -p && git commit.

Where next

• Coding in Mel — the full terminal + agent + editor + review loop on a real task.
• Keyboard shortcuts — everything you just used, plus the rest. Or press Ctrl+Shift+P (⌃P on a Mac) and search.