Mel: The Agentic Terminal for Running a Team of AI Coding Agents

Most AI coding tools make you pick a side: a chat window, an editor plugin, or a single CLI agent. The Mel agentic terminal takes a different angle. It’s a desktop terminal where a team of AI agents works on your files, commands and code, and you stay in the loop through approvals you configure yourself. It runs on Mac, Windows and Linux, there’s a VS Code extension, and it works with Claude, GPT, Gemini, DeepSeek, Qwen and more.

This is a look at what Mel is, how it works, what it costs, and who it suits. I haven’t run it on a large production codebase, so this is based on the project’s own documentation and feature list rather than a long-term review.

What Mel is

Mel is built by Concepts Dzign LLC and Thinkverse AI LLC, and it’s written in Rust. The pitch is a terminal-native workspace: agents read files, run commands and edit code inside the terminal, so you aren’t bouncing between a chat app, a shell and an editor.

It’s a different thing from a terminal emulator like iTerm2 or Ghostty, which render a shell and leave the thinking to you. It’s also different from a single agent like Claude Code. Mel sits somewhere in between: a terminal app whose built-in agents plan work, run commands, check the results and report back.

Key features

• Plan-first workflow: Mel plans the task, runs the commands, checks the outcome, and can fall back to another provider if one fails.
• Multi-agent orchestration: you can run a team of Mel agents in parallel from the terminal rather than one conversation at a time.
• Built-in code review: changed files show up as diffs in a review panel, and an agent can review a branch or pull request locally for bugs and security issues.
• Third-party agents as teammates: Mel can run Claude Code and Codex CLI alongside its own agents, each in its own pane.
• Integrations: SSH sessions, MCP servers, project memory, skills, a VS Code extension, and Telegram notifications. The site lists 17 connectors, each used with your own account or key.
• Beyond code: the same agents can tidy folders, build sites and automations, and turn finished work into shareable PDF reports.

Models it works with

Mel is deliberately model-agnostic. It supports Claude, GPT, Gemini, DeepSeek, Qwen, Kimi and GLM, with others such as Grok and MiniMax also listed. An auto mode picks a model for each task, or you can pin a specific one. For anyone who switches between providers depending on cost and quality, that’s the main draw: one terminal, one set of approvals, any model behind it.

Approvals: staying in control

Letting an agent run shell commands is where most of the risk sits, so the approval model matters. In Mel you configure it. You can require confirmation for file writes, edits and commands, or allow specific actions to run on their own once you trust them. A sensible starting point is to approve everything, then loosen reads and test runs first while keeping writes and anything destructive behind a prompt.

Platforms and installation

• macOS: Apple Silicon and Intel, macOS 11 or later. Download the signed, notarized zip and drag Mel to Applications.
• Windows: Windows 10 and 11, with an installer or portable zip, including a native ARM64 build.
• Linux: x86_64. Extract the .tar.gz and run ./mel. It needs OpenGL.
• VS Code: install the extension with code --install-extension Mel.openmel. It needs a recent Mel app (0.1.48 or newer).

Pricing

Mel is free to download and use with your own API key; you pay your model provider directly. Paid plans add hosted credits so you don’t need to manage keys:

• Free: bring your own provider key.
• Pro: $20 a month.
• Max: $200 a month.
• Business: $50 per seat per month, two seats minimum, with a shared credit balance, invoicing and admin controls.

Things to know before you try it

• It’s young. The download page showed v0.1.64 when I looked, so expect fast change and some rough edges.
• It isn’t open source yet. The site says open source is coming but not available, so you can’t audit the code today.
• Test it on a throwaway repo first and keep approvals tight until you’ve seen how it behaves.

Where it fits

If you want a fast, plain terminal, stay with something like Ghostty or iTerm2; see my guide to the best terminal emulators for developers. If you want AI built into the input bar, Warp is the closest comparison. If you already like running agents in git worktrees, Orca covers similar ground as a control layer above your existing agents.

Mel is worth a look if you want one desktop app that combines a terminal, a team of agents, any model you like and explicit approval controls, and you’re happy to use a very new tool to get it.