AI Devops
-
Gotta say, Claude Opus 4.5 does feel pretty close to AGI for development and dev-ops.
If you haven't tried it yet, highly recommended, especially as it can ssh into your cloudron instances, answer questions, and help with maintenance, optimisation, and even migrations.
I've spent the last couple of weeks creating this repo, and now website, as an outlet for all the agents I can now create and put to work with it:
- https://github.com/marcusquinn/aidevops (appreciate a star)
- https://aidevops.sh/ (created with Opus 4.5)
My round the clock stack is now:
- https://tabby.sh/ (I like with Tabs on the left, and using Profiles per repo)
- https://opencode.ai/ (just a beautiful TUI, it makes all the others feel lame)
- https://zed.dev/ (so fast and clean compared to VSCode/VSCodium)
With all this, I'm at the point where nothing feels impossible, and years of ideas and ambitions can now just be a Tabby tab conversation with Opus via OpenCode away from getting anything done.
All open-source, so you can use AI to ask what it all does, if it's safe, etc.
Giving it hosting access and Cloudron instances to play with is unlocking infinite potential — that we could never do with closed SaaS services.
To the creators and builders, our time is now
Web Design & Development: https://www.evergreen.je
Technology & Apps: https://www.marcusquinn.com -
Gotta say, Claude Opus 4.5 does feel pretty close to AGI for development and dev-ops.
If you haven't tried it yet, highly recommended, especially as it can ssh into your cloudron instances, answer questions, and help with maintenance, optimisation, and even migrations.
I've spent the last couple of weeks creating this repo, and now website, as an outlet for all the agents I can now create and put to work with it:
- https://github.com/marcusquinn/aidevops (appreciate a star)
- https://aidevops.sh/ (created with Opus 4.5)
My round the clock stack is now:
- https://tabby.sh/ (I like with Tabs on the left, and using Profiles per repo)
- https://opencode.ai/ (just a beautiful TUI, it makes all the others feel lame)
- https://zed.dev/ (so fast and clean compared to VSCode/VSCodium)
With all this, I'm at the point where nothing feels impossible, and years of ideas and ambitions can now just be a Tabby tab conversation with Opus via OpenCode away from getting anything done.
All open-source, so you can use AI to ask what it all does, if it's safe, etc.
Giving it hosting access and Cloudron instances to play with is unlocking infinite potential — that we could never do with closed SaaS services.
To the creators and builders, our time is now
Oh, you can use Opus 4.5 for free on Google's Antigravity at the mo.
Obviously not for this setup, but a good way to give it a try on general tasks and projects.
-
Gotta say, Claude Opus 4.5 does feel pretty close to AGI for development and dev-ops.
If you haven't tried it yet, highly recommended, especially as it can ssh into your cloudron instances, answer questions, and help with maintenance, optimisation, and even migrations.
I've spent the last couple of weeks creating this repo, and now website, as an outlet for all the agents I can now create and put to work with it:
- https://github.com/marcusquinn/aidevops (appreciate a star)
- https://aidevops.sh/ (created with Opus 4.5)
My round the clock stack is now:
- https://tabby.sh/ (I like with Tabs on the left, and using Profiles per repo)
- https://opencode.ai/ (just a beautiful TUI, it makes all the others feel lame)
- https://zed.dev/ (so fast and clean compared to VSCode/VSCodium)
With all this, I'm at the point where nothing feels impossible, and years of ideas and ambitions can now just be a Tabby tab conversation with Opus via OpenCode away from getting anything done.
All open-source, so you can use AI to ask what it all does, if it's safe, etc.
Giving it hosting access and Cloudron instances to play with is unlocking infinite potential — that we could never do with closed SaaS services.
To the creators and builders, our time is now
@marcusquinn Wow that's great stuff man thanks for sharing.
Release 2.13.0 already, why have you waited so long to share it you don't seem to be the shy kind?
-
@marcusquinn Wow that's great stuff man thanks for sharing.
Release 2.13.0 already, why have you waited so long to share it you don't seem to be the shy kind?@micmc aha - i'm iterating fast and pretty-much daily
What you're getting, is what I'm using every day now, so it's all being refined from real-life use, and self-improving using the @agent-review subagent.
I've added a ton for development workflow best-practice with git, to help anyone vibe-coding stay better organised and safer with their workflow.
Issues and PRs welcome. It is all opinionated for the tools i work with most, but hopefully all self-explanatory enough to see how to extend for your own needs.
I'm having a ton of fun with it. Hope you all enjoy the journey as much as I am. It really is becoming the dream team
-
Looks really interesting!
Quick question: Do you find yourself paying for Claude Max, or Claude Pro, or just use API calls only for a pay-as-you-go (not necessarily for this project but just in general)? I was a ChatGPT Plus subscriber and while it has some nice features that Claude doesn’t have which make my life a bit easier compared to if I only used Claude web/desktop, in my testing I do enjoy Claude’s answers so much better in most cases. But the limits they have on their Pro plan I run into constantly so I think I need Max but it’s hard to justify that significant increase in cost. I wish they had something in the middle.
--
Dustin Dauncey
www.d19.ca -
Looks really interesting!
Quick question: Do you find yourself paying for Claude Max, or Claude Pro, or just use API calls only for a pay-as-you-go (not necessarily for this project but just in general)? I was a ChatGPT Plus subscriber and while it has some nice features that Claude doesn’t have which make my life a bit easier compared to if I only used Claude web/desktop, in my testing I do enjoy Claude’s answers so much better in most cases. But the limits they have on their Pro plan I run into constantly so I think I need Max but it’s hard to justify that significant increase in cost. I wish they had something in the middle.
@d19dotca Thanks. I'm on Claude Max ($200/month) at the mo, and using heavily most days.
Opus 4.5 ain't cheap with the API, but I think you can get by with Sonnet or Gemini for most general tasks. I just need that extra architecture intelligence for developing anything from scratch.
I think they are all generalists, though. What I need is more controlled design patterns, and less repeating information that's specific to my infrastructure. Hence this repo.
I think this kind of setup is going to be the main differentiator in how much value anyone can get from AI, as the generalist abilities are fine, but just too time-consuming to keep repeating yourself inconsistently with.
The real leverage is this kind of progressive knowledge pattern of main agents and subagents.
That's the best explainer I can give in a few words, the full explainer is just the repo itself. Like any good codebase (and clear English is now a programming language), the best systems are ones that are self-explanatory.
Yeah, if you're not making money from it directly, it's a tough balance. I think in that case API usage and a lot of discipline on token usage is needed, which again this sort of setup is designed to continually do.
I think the winners in AI are going to be those that can compress and manage context efficiently and with clear organisation.
Hope it's at least given you some inspiration. As much as I'm happy with the tools I'm using (above), I have tried to follow standards for naming and organisation, so it can be used with any other AI interfaces. I just only have time to test what I use.
