Sign in

Atlarix

The private AI workstation for engineering teams.

macOSWindowsFree

Listed by abishaiama,

Leave feedback

Atlarix is an AI coding workstation: a desktop app for macOS, Windows and Linux, plus a terminal CLI and a VS Code extension. Open a project folder, say what you want built or fixed, and the agent reads the code, makes the changes and runs the commands, asking before anything risky. Use the free Auto model, or bring your own API key.

What to test

Download the desktop app from atlarix.dev, sign in and open one of your own projects. Ask it to explain the code, then to fix a small bug or add a test, and review the changes it makes.

Bounty
?per approved post

The bounty is what this listing pays a tester per approved post. Sign in to see it.

Download page, opens in a new tabWebsite, opens in a new tab

Completely free.

This app is made and run by its builder, not by TestMyBuild. We tried it once before listing it, which is not a check that it is safe, keeps working or handles your data well. Use your own judgement.

Report listing

Feedback

1 tester, on Windows

1 post, newest first
jan
Tested on Windows
Signup screenshot

Started on your website to install it. The download flow is solid: Windows goes through the Microsoft Store and macOS is signed and notarized. That's how a desktop app should be distributed, and why we allow it on TestMyBuild (we don't permit direct download links).

1. Download from atlarix.dev

  • The main CTA is very clear, on mobile too.
  • Missing trust signal: contact details. No address, email or business info in the footer. I see it's built by Nora Labs, but who do I contact if something goes wrong? The app runs locally and touches API keys, so show who's behind it: address, registration, phone, email. It says "We're confident it's safe, and this is who we are." We do the same on TestMyBuild.
  • Your audience is developers, so show the terminal command to verify the macOS signature.
  • "Talk to Us Directly" opens an email. Link to a contact section instead.
  • Too many nav items. I'd keep about four. Not urgent, since the main flow (land → download) works.

2. Sign in

  • Google, GitHub, Discord and email are good options.
  • "More about sign in" feels unnecessary.
  • The browser auth page stays open after signing in with Google.

3. Open a project

  • Picking a workspace is familiar, like VS Code.
  • Switching projects means closing the app. The workspace name in the sidebar looks clickable but does nothing. Perfect spot for a VS Code-style switcher.
  • The home page is called "Apps", which is an odd name.

4. Explain the code

  • Worked really well. In ~20 seconds it correctly described my stack (Astro on Cloudflare Workers), folder structure and deploy flow.
  • The modes (Explore, Plan, Build, Debug, Review) are a great idea and clearly explained.
  • Big gaps between list items make short answers long to scroll.
  • On a more complex feature I wanted to build, it didn't think very long. Fine, but worth knowing.

5. Add a test
I asked for one unit test for a simple helper, without touching existing files, and to run it.

  • It picked a sensible function to test.
  • Its first test wouldn't have run. It used `describe`/`it` without importing them and imported a `.ts` file from a `.mjs` test. It only fixed this after reading my existing tests. It should check how the project runs tests first.
  • It never confirmed the test passed. After ~8 minutes and 3 rewrites the run was still hanging, with 6 background tasks left running that I had to ask it to stop. It did honestly say it couldn't claim it finished, which I appreciate.
  • Internal errors (_"each edit needs old_content and new_content"_) show up in the chat, and some "Thoughts" blocks have reply text glued into them.

6. Review the changes

  • A real strength. Clean diffs with a summary, per-hunk approval, and fine-grained "always allow" options (file, folder, file type).
  • "Asks before anything risky" holds up. Deleting a file was flagged DANGEROUS with a clear warning.
  • The Revert button on each answer is nice.
  • The approval bar covers "Jump to latest", so it's hard to follow the agent while it waits for me.

Models & pricing

  • Running locally feels safer, and the free Auto model is a great trust builder. Paying for other models makes total sense, and I like being able to switch.
  • The free-usage meter is confusing. After two prompts it went from ~12% running low to ~0% to ~100%. Used or remaining? Am I about to be cut off?
  • The Auto model can't see images, so computer use is off with it. Say that upfront.
  • The Tools section is unclear. Is a tool a skill, does it work with skills, or is it a connector?

Feature wishes

  • Agent-to-agent view. I connect Codex and Claude via MCP but can't see what they say to each other. Showing that live while I code would be a killer feature for me.
  • Design previews, like Claude's artifacts, before it builds.
  • Voice input. I use Wispr Flow, but it'd be a nice addition.

Biggest takeaway: positioning
Your funnel got me to download without even asking what it does, but I still don't know what Atlarix is. A terminal? A coding agent? Its own LLM? The site copy also reads as AI-written (dashes, "waiting for your approval"), which hurts credibility with developers.
For someone already using Claude or ChatGPT, or comparing tools right now: what do you do that they don't? Why would I use Atlarix instead? Your approval and diff system is a strong answer, so lead with it. That's the main thing to work on.

from the team