Hi @Joseph! We have definitely experimented with this internally and have gotten some proof of concepts working in the past, so I think it would be great to hear about and learn from your experiences in getting this setup! I’m not personally aware of anyone in the community that has attempted this but if so it’d be great to hear from them as well. FWIW we focused on using tasks that submitted work to a docker daemon and waited / reported state as needed, but I’d be curious to know what setup makes sense in your case.