Skip to content
New — orchestrate a team of Mel agents from your terminal. Learn more about orchestrating Mel agents
mel

Mel – the agentic terminal
The AI terminal for teams.

Mel Red Team · beta

A red team of models that reviews your code.

Run mel redteam in a project. Models from different companies each review a snapshot of your code, check one another's points, and Mel writes one plain report. It runs nothing.

In the beta version of Mel (v0.1.63), from the command line.

How it works

What happens when you run it

  1. 01
    Run it in your project

    Mel copies the committed code into a snapshot, with no git history, links or submodules. Your working copy is not touched. Add --base to review only the change since a branch. The program is inside the Mel app; on a Mac it is /Applications/Mel.app/Contents/MacOS/mel.

    mel redteam
    mel redteam --base main
  2. 02
    Each model reviews alone

    Each model on the panel reviews the snapshot on its own, with reading tools only. No model sees another model's words.

  3. 03
    The models check one another

    Each model then sees the points it did not raise, with no author named, and answers confirm, reject or unsure, with a line of evidence. This step has no tools.

  4. 04
    One report

    Mel merges the points that match and writes report.md and report.json.

Review only

It reviews and reports. It runs nothing.

No commands run

This version runs no commands: not your tests, not a linter, and not a command a model suggests.

Commands stay text

A command a model suggests is shown as text, marked "Not run (this version runs no commands)". A point with no command is marked Opinion.

Nothing called proven

Nothing in the report says a problem was reproduced, because nothing was run.

Your account stays clean

The review is not filed in your conversation list or uploaded to your account. Its temporary files are removed when the run ends.

The panel

One model per company

Cheap panel

The default: three low-cost models, one each from Anthropic, OpenAI and Google.

Premium panel

Four top models, one each from Anthropic, OpenAI, Google and xAI.

Your own panel

Pick 2 to 6 models with --models. Mel refuses two models from the same company.

A credit limit

A run starts no new step once 300 credits are spent. Set your own with --max-credits.

Coming soon

Coming soon

  • Proof and checks: coming soon. Run your project's own checks and each suggested command in a sandbox with no network, and see which points were reproduced.

Get started

Have several models review your code.

Mel Red Team is in the beta version of Mel. The terminal is free to download.