introducing agent-loops + ui viewer
@bentossell203K followers
117,023 views
22 Comments
Comments from the original post on X (Twitter)

@AshwaryaMaratha
Thanks for sharing this. @mattpocockuk 's work has still not received enough attention. I actually love the way he has figured out things.
5 likes

@pedronauck
Give a try on Compozy, you will love it! It's built on top of Claude Code (you can use your own subscription), we have a strong PRD->Techspec->Tasks brainstorming process, our looper is very optimized, support for worktree natively!
5 likes

@bentossell
you should use @droid then i may try it
2 likes

@pedronauck
@droid it's in our roadmap to implement Droid as provider as well, maybe we could talk to speed up this? what do you guys think?
1 like

@bentossell
@droid what do you need? we have headless mode

@pedronauck
@droid I sent you a DM Ben, I guess we can have a nice opportunity for you guys here in Brazil to hype Droid not just on Compozy, but in our AI course that we have more than 1000+ students now 🙏🏻

@tomosman
ledge
5 likes

@coardmiller
Out of curiosity, how does this differ from I have been exploring that and trying to hone in on the best solution
3 likes

@bentossell
haven’t used it tbh. i just wanted to build my own system
1 like

@Timur_Yessenov
github issues as the trigger layer is smart. most people overcomplicate the orchestration part when the simplest interface wins. the ui viewer piece is underrated too - debugging blind loops is painful.
3 likes

@searchers_com
Nice work 🥳🥳🙏🏼
2 likes

@bentossell
ty
1 like

@mysticaltech
Awesome, would be better to use a monorepo for both. Thanks for sharing
1 like

@bentossell
i built them at separate times and the loop is plug and play to lots of things. but noted ty!
3 likes

@notcalmerecentl
, Vibe kanban, Autoclaude
1 like

@JBabaTalks
I like it let me try it
1 like

@nbbaier
Seems like @dannypostma's original tweet is gone what was it?

@auster_felix
Great set-up. Quick qn: in Droid I’d like to have GPT 5.2 (xhigh) for planning and GPT 5.2 Codex (xhigh) for coding. Somehow whenever I run /model to choose spec mode model I can only get GPT 5.2 (low) with no other choice. Bug or me being stupid?

@bentossell
sounds like a bug to me will check
1 like

@Flowgrammers
Impressive work tying agent-loops into a UI viewer and cross-environment loop execution from local/GH. How are you handling auth and loop state consistency across endpoints? Any plans to publish a protocol or guide?

