What if Claude Code, Codex, Cursor knew how to do your work without you explaining it? AgentHandover observes you, learns, and teaches your agents with self-improving Skills.
No reviews yetBe the first to leave a review for AgentHandover
Maker
📌
So you know how every time you use OpenClaw, Claude or Codex you end up spending 20 mins explaining how to do something you could do in 2? And then next time you have to explain it all over again?
I built this mac app/CLI that just watches you work quietly. it figures out what you did, asks you a couple questions like "did you actually want to send that or save as draft?" when it needs to, and then any agent you connect just knows how to do it.
The coolest part is when the agent runs it and does something slightly different, the skill rewrites itself. It runs fully local, nothing leaves your machine. open source.
What would you hand over first?
Report
How does it know if I am doing work or just doomscrolling on my laptop lol?
Report
Maker
@pedro_henrique6759 it's a multi-stage pipeline - activity classifier filters obvious noise, then the local AI analyzes what you're actually doing on screen, then task boundary detection groups coherent work, then pattern detection only promotes repeated workflows to skills. random browsing dies at stage 1, one-off tasks die at stage 4.
Report
if this is local-first that’s a big plus, don’t want all workflows going to the cloud
Report
Maker
@xavair its local!!!!!!! u can do cloud and APIs but we don't recommend as there is no need
Report
Is it Qwen only or any model?
Report
Maker
@john_keegan1121 we recommend Qwen 3.5 but you can choose any Ollama model or you can also do API cloud way but there is no need for it
Report
This is actually interesting. The biggest pain with agents right now is repeating context every time. If this really learns workflows over time, that’s a big shift.
Report
Maker
@moh_codokiai thank you, it should! It will watch learn analyse write and rewrite - also we each executed workflow it will improve
Report
I’d probably use this just to avoid writing the same instructions 10 times a day 😅
Report
Maker
@josh_bennett1 haha, that was the starting motivation for me to create it! :D
So you know how every time you use OpenClaw, Claude or Codex you end up spending 20 mins explaining how to do something you could do in 2?
And then next time you have to explain it all over again?
I built this mac app/CLI that just watches you work quietly. it figures out what you did, asks you a couple questions like "did you actually want to send that or save as draft?" when it needs to, and then any agent you connect just knows how to do it.
The coolest part is when the agent runs it and does something slightly different, the skill rewrites itself. It runs fully local, nothing leaves your machine. open source.
What would you hand over first?
How does it know if I am doing work or just doomscrolling on my laptop lol?
@pedro_henrique6759 it's a multi-stage pipeline - activity classifier filters obvious noise, then the local AI analyzes what you're actually doing on screen, then task boundary detection groups coherent work, then pattern detection only promotes repeated workflows to skills. random browsing dies at stage 1, one-off tasks die at stage 4.
if this is local-first that’s a big plus, don’t want all workflows going to the cloud
@xavair its local!!!!!!! u can do cloud and APIs but we don't recommend as there is no need
Is it Qwen only or any model?
@john_keegan1121 we recommend Qwen 3.5 but you can choose any Ollama model or you can also do API cloud way but there is no need for it
This is actually interesting. The biggest pain with agents right now is repeating context every time. If this really learns workflows over time, that’s a big shift.
@moh_codokiai thank you, it should! It will watch learn analyse write and rewrite - also we each executed workflow it will improve
I’d probably use this just to avoid writing the same instructions 10 times a day 😅
@josh_bennett1 haha, that was the starting motivation for me to create it! :D