What is a terminal for AI coding agents?

A terminal for AI coding agents is a terminal app built around agents, not only around commands. It runs CLI agents such as Claude Code and Codex in their own panes, tells you which agent is waiting for you, and can run its own team of agents under approvals you set.

Mel – the agentic terminal is this kind of terminal. Any terminal can start a coding agent: you type claude or codex and it runs. The difference shows up when you run three of them at once, leave one in a hidden tab, and come back an hour later to find it has been waiting for a yes the whole time.

What a terminal for AI coding agents should do

1. Run the agents you already use. CLI agents take over the screen, so the terminal has to render them properly and route your keys to them.
2. Keep many agents in view. Tabs, split panes and a sidebar that shows which session is running and which is blocked.
3. Say when an agent needs you. An agent asking for approval in a tab you are not looking at should not wait in silence.
4. Keep agents apart. Two agents editing the same folder will undo each other's work, so each needs its own branch or worktree.
5. Offer its own agents too. A team of agents that plans, builds and tests one task, with no setup.
6. Keep you in control. Approvals for commands and edits, and a clear view of every change.

How Mel does it

• CLI agents run as live sessions. Claude Code, Codex CLI, Gemini CLI, GitHub Copilot CLI, OpenCode and Amp are recognised when you start them and shown full-pane, with a tool belt at the bottom. When one exits, its last screen folds into a normal block.
• A sidebar of every session. Vertical tabs list every tab and every split pane, each with a live status: running or blocked.
• Notifications when an agent needs you. When an agent in another tab is blocked on an approval or finishes its turn, Mel raises a card and adds it to the inbox. A click takes you to that tab.
• One screen for all runs. The Runs screen groups every agent run in three lists: Needs you, Running and Recent.
• Worktrees in two clicks. Track a repo from the + menu, then create a worktree on a new branch, one agent per worktree tab.
• Mel's own team of agents. /orchestrate-agent splits a task between a lead and workers, each in its own pane. /orchestrate-cli makes Claude Code or Codex the workers.

How the options compare

Tips for running coding agents in a terminal

• Start from a clean git tree, so the diff shows only the agent's work.
• One agent per worktree when tasks run side by side.
• Give each agent a finish line, such as "npm test passes".
• Read the diff, not the summary. Mel's review panel shows every change in one place.

FAQ

What is a terminal for AI coding agents? A terminal app built around agents: it runs CLI agents in panes, shows which one is waiting for you, and keeps every agent under your approvals.

Can I run Claude Code and Codex in Mel? Yes. Type claude or codex in a Mel pane and it runs as a live session. You need those tools installed and signed in.

How do I know when an agent needs my approval? Mel shows a blocked status in the sidebar, raises a notification and adds it to the inbox. The Runs screen lists every run that needs you.

Do I have to pay to use Mel with my own agents? No. Mel is free with your own API key on Mac, Windows and Linux. Claude Code and Codex are billed by their own makers.

• How to run several AI agents at once
• What is a multi-agent terminal?
• How to manage AI coding agents
• Mel vs Claude Code
• Mel vs Codex CLI