Ixel
Ixel · To shine

Put your AI models to work together.

Ixel makes small, local tools for people who use more than one AI. Have Claude, GPT, Gemini and local models check each other's answers. Let Claude and Codex hand work back and forth. Reach the agents on your servers in one click.

  • macOS · Linux · Windows
  • Runs on your computer
  • Your keys stay with you
Three stars in the Ixel mark, one for each tool
The tools

One for getting a verdict, one for sharing the work, one for reaching your agents.

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 agree on a verdict.

  • Answers are shuffled and labeled A, B, C for each reviewer, and a model's grade of its own answer doesn't count. You see who was flagged, and who conceded.
  • Usage saver mode lets cheaper and local models draft and review, while your best model only checks the result.
  • Uses the keys and subscriptions you already have: Claude Code, Codex, Gemini CLI, GitHub Copilot, OpenCode, and local models through Ollama or LM Studio.
  • Terminal
  • Browser app
  • Plugin for Claude, Codex, Cursor
  • MIT
More about Ixel MAT →
● 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
β Agents as partners

Handoff

Let Claude and Codex work as partners on the same project. Say "Codex, take the tests; Claude, do the API," and the work passes between them. You stop being the copy-paste courier.

  • A plugin for Claude Desktop, Claude Code and Codex. Agents create tasks, hand them over with notes and acceptance checks, and ask each other for reviews.
  • The board enforces its own rules: only the assignee can hand a task on, nobody reviews their own work, and the history is append-only.
  • You stay in charge from the terminal. handoff board shows who's doing what, what's blocked, and everything that happened.
  • Claude Desktop
  • Claude Code
  • Codex
  • MIT
More about Handoff →
● handoff boardExample
handoff board
Handoff board  3 active · 1 open · 1 claimed · 1 handed_off

Task  Status      Assignee  Title
T-2   handed_off  claude    Tests for /api/orders
T-1   claimed     claude    Add /api/orders
T-3   open        codex     Update the API docs
Then, in Codex Check your handoffs.
γ SSH for AI agents

Ixel Console

The SSH launcher built for the age of AI agents. Save your machines as profiles, pick the agent on each, and connect in one click, from a $5 VPS to your home lab.

  • Works with OpenClaw, Hermes, or any agent command you type. Each host keeps its own agent and command presets.
  • Tabbed sessions, one command broadcast to every session, and one-click Status, Sessions, Logs and TUI macros.
  • Host keys are pinned, and a mismatch blocks the connection. Credentials live in your system keychain, and every connection attempt is logged on your machine.
  • Linux
  • macOS
  • Windows / WSL2
  • MIT
More about Ixel Console →
● profilesExample
HostAgentHost key
homelabOpenClaw✓ pinned
vps-fra1Hermes✓ pinned
gpu-boxcustom✓ pinned
StatusSessionsLogsTUIBroadcast to all
How we build

Built the same careful way, all three.

Every Ixel project ships a written threat model and tests that attack its own protections. Here is what that means for each tool.

αIxel MAT βHandoff γIxel Console
Stays on your machine The browser app listens on 127.0.0.1 only and needs a key from the link it opens. Never opens a network port. Apps talk to it over stdio. No telemetry. It connects only to the sessions you start.
Keeps secrets out of reach API keys live in a file only you can read, and go only to the providers you set up. Text that looks like an API key, token or private key is refused, not stored. Credentials live in your system keychain, never in profile files.
Limits what can run Subscription CLIs run answer-only: no shell, no file writes, in an empty folder. Runs nothing itself. The optional worker stays in its own git worktree and never pushes. Fixed commands per agent. Custom commands are length-capped and checked for shell tricks.
Attacked in its own tests A fake model tells each real CLI to run a command and read the key file. CI fails if anything happens. Tests include a crafted board built to hijack an agent, and an attack on the worker. SSH option-injection tests run through the real ssh binary.
About Ixel

We help people get more from the AI they already use.