Ixel
α Multi-agent terminal

Ixel MAT

Ask several AI models at once. They answer, grade each other's work without knowing whose it is, and one of them writes the verdict.

Ixel MAT puts the models you already use on one panel: Claude, GPT, Gemini, Grok, local models through Ollama or LM Studio, and your Claude Code, Codex, Gemini CLI, Copilot, OpenCode and Grok Build logins. Use it in your terminal, in your browser, or as a plugin inside Claude Desktop, Claude Code, Codex and Cursor.

The α star in the Ixel mark
See it run

Ixel MAT in a minute.

● ixel reviewExample
ixel review "What is 17 × 23?"
Round 1Answer
Round 2Peer review
Round 3Verdict
IXEL VERDICTconfidence high
17 × 23 = 391
Corrections: one answer said 381.
AGemini0.00✗ flagged by 2
BGPT1.00
CClaude1.00
↪ Gemini conceded: rated C above its own answer
Example run. The recording plays here when it loads.
A real run: Claude (Sonnet), Grok Build and Qwen3 4B (running on the computer itself) answer, grade each other, and Claude writes the verdict. Long waits are sped up. All the videos
Who it's for

Made for people who…

  • Need an answer they can trust

    Ask about a regex, a query or a tricky bug, and see where the models disagree before you rely on the answer.

  • Pay for more than one AI

    Put your Claude, ChatGPT, Gemini and Copilot plans on one panel instead of asking each one separately.

  • Watch what AI costs

    Saver mode lets cheaper and local models draft while your best model only checks, and every review ends with what it cost.

  • Are about to push code

    Hand the panel your diff before you push. It looks for bugs, security problems and anything that would break. It only reads the code: API models and the built-in CLI setups get no file or shell access (an agent you add yourself, such as an OpenClaw gateway, keeps whatever tools it has).

How it works

Three rounds to a verdict.

  1. Answer

    Every model on the panel answers the question on its own.

  2. Peer review

    Each model grades all the answers. They're labeled A, B, C in a random order, and each reviewer sees them in a different order, so nobody knows whose work they're grading. A model's grade of its own answer doesn't count.

  3. Verdict

    A moderator writes the final answer from the answers and the reviews, and says where the panel disagreed. You see who was flagged and who conceded.

What it does

What you get.

Four modes, or Auto
Quick, Review, Deep and Saver. With triage turned on, Auto picks how much checking each question needs, so a simple one doesn't pay for peer review.
Review your code
Your uncommitted changes, what you've staged, a whole branch, or files you pick. Common kinds of API keys, tokens and private keys, and files such as .env, are refused before they're sent. Passwords aren't caught.
Usage saver
Cheaper models draft and review, and your big model confirms a draft or sends a wrong one back, usually in a few words instead of a whole answer. Saves, streaks and achievements keep score, and ixel saves shows about how many big-model tokens it saved, and how many dollars when that model is on an API key.
What every review cost
Dollars on API keys (prices for Claude models are built in; add your own for other providers), calls on your subscriptions, and free runs on your own hardware.
Three ways in
The terminal (ixel), a browser app only your computer can open (ixel gui), and a plugin for Claude Desktop, Claude Code, Codex and Cursor.
The newest models
For API models (OpenAI, Anthropic, Gemini and xAI), set a model to latest and Ixel uses that provider's newest top model, looked up when it starts. A model with a new kind of name needs an Ixel update first.
Built to be safe
  • Your keys stay on your computer, encrypted where there's a system keychain, and your questions go only to the providers you set up.
  • Subscription CLIs run answer-only: no shell, no file writes, in an empty folder. Gemini CLI can still search and read the web. A test, run by hand, attacks each one to check this.
  • Model output is treated as untrusted. Answers are shown as text and fenced off before another model reads them.
  • The browser app listens only on your computer and needs a one-time key.
Install

Get Ixel MAT running.

You need:

  • Windows 10 or 11, macOS 12 or newer, or Linux
  • Python 3.10 or newer (the installer offers to get it, asking first). On Debian and Ubuntu, also python3-venv
  • Git (the installer offers to get it if it's missing)
  • At least one AI: an API key, a subscription CLI such as Claude Code or Codex, or a local model

This installs Ixel MAT and the Ixel app. It downloads Ixel MAT from GitHub and updates it when you run it again. The installer itself needs no administrator rights (getting Git or Python may). You can read it first: install.ps1, install.sh.

Windows: in a normal PowerShell window (not "Run as administrator"), run:

irm https://ixelai.com/ixel-mat/install.ps1 | iex

macOS or Linux: in a terminal, run:

curl -fsSL https://ixelai.com/ixel-mat/install.sh | sh

Then, in a new terminal window, run ixel setup to choose the models on your panel. With a shell other than bash, zsh or fish, the installer first prints the line to add to your PATH.

Or get everything with Ixel

Ixel installs Ixel MAT, Handoff and the Ixel app together. Handoff lets your coding agents hand work to each other and ask your Ixel MAT panel for reviews. Already have Ixel MAT? Type /handoff in the Ixel app to see how to add Handoff.

Windows:

irm https://ixelai.com/install.ps1 | iex

macOS or Linux:

curl -fsSL https://ixelai.com/install.sh | sh