←
windischb107 posts
windischb avatar

windischb

@windischb4
◷ Joined March 2019▧ 17 Photos and videos
windischb avatar
windischb@windischb4·3h

Got the chance to test @MistralAI for the next three months — time to put it to work in @codealta ! 🧪 Looking forward to going beyond the benchmarks and seeing how the models hold up in my day-to-day development workflow.

Post attachment
windischb avatar
windischb@windischb4·Nov 13

Huge props to @xoofx for building CodeAlta. This was a pretty wild way to test a PR. This is crazy. And honestly? Pretty amazing. Did I mention crazy? 🤯

windischb avatar
windischb@windischb4·Nov 13

All of this ran in CodeAlta’s Orchestrator mode from one prompt, following a previously agreed plan. upcoming prompt: update the issue and create the PR. What amazes me: AI didn’t just write the code. It built the environment needed to test it. 🚀

windischb avatar
windischb@windischb4·Nov 13

Then came the test cycle: Run tests → review results → fix a small bug → copy changes into the VMs → rebuild → restart → test again. Once everything worked as agreed in the plan, CodeAlta produced a report with the test results.

windischb avatar
windischb@windischb4·Nov 13

CodeAlta’s Fleet feature can also install and configure local LLMs. So the agent installed llama.cpp, set up the models, and configured the CodeAlta instances to use them. The test fleet now had its own local AI available for the tests.

windischb avatar
windischb@windischb4·Nov 13

Another requirement: the instances inside the VMs needed to run AI sessions during testing. I didn’t want to sign into my Claude Code or Codex accounts on those machines. The solution: local LLMs. And yes, the agent handled that setup too.

windischb avatar
windischb@windischb4·Nov 13

Then it copied the source into the VMs, compiled CodeAlta for each platform, launched it, and configured the fleet connections. The result: One Fleet Server + three connected CodeAlta instances across macOS, Windows, and Ubuntu.

windischb avatar
windischb@windischb4·Nov 13

Using VMPal by @TablePlus - a VM app controlled through MCP - the agent created three VMs: → Windows → Ubuntu with a desktop → Headless Ubuntu for the Fleet Server It installed the operating systems, dependencies, and frameworks. No manual VM setup.

windischb avatar
windischb@windischb4·Nov 13

My installed version of CodeAlta didn’t even include the changes yet. So CodeAlta built and launched a separate dev instance from its own source code - with my uncommitted changes. Next: give that instance some machines to connect to.

windischb avatar
windischb@windischb4·Nov 13

One prompt. Three operating systems. A complete test fleet. 🚀 I’m working on a small improvement for @codealta’s Fleet feature: fetching prompt suggestions from remote instances. To test the PR, I needed multiple machines. My starting point? Just my Mac.

windischb avatar
windischb@windischb4·Nov 11

And it's not Electron. It's C# on .NET 11, released just yesterday – a web UI inside a real .NET app. Made possible by NeoAstra, the open-source framework by @xoofx. Thank you, Alexandre! 🙏 github.com/NeoAstra/NeoAstra

windischb avatar
windischb@windischb4·Nov 11

So I moved everything onto one shared core: iPhone, iPad, Android and Fire tablets today – Windows, macOS and Linux next. Still native where it matters: passkeys, push, store purchases, and a cooking timer with a Live Activity on the iPhone lock screen.

windischb avatar
windischb@windischb4·Nov 11

amZettel – shopping lists, family calendar, recipes, meal plans – has been on iPhone for a while. Shipping the native Android app would have meant two apps: every feature built twice, and Android users always waiting a few weeks.

windischb avatar
windischb@windischb4·Nov 11

I had a finished Android app. I never released it. Instead, I rebuilt amZettel from the ground up – and now every new feature ships on iPhone, Android and Fire tablets on the same day. 🧵 #Dotnet

Post attachment