https://github.com/OpenHands/software-agent-sdk/pull/2206 - implementing ability to hard terminate the agent (even mid llm response) would love early feedback; using acompletion from litellm but running it in a separate thread to preserve existing sync api scheme (instead of having to go async everywhere) there is an example script that demonstrates the immediate interrupt ability maybe cc @straight-sundown-84319

Makes sense - we introduce the async only at the agent server level

Would be great if you left some context regarding the sync only api scheme: https://github.com/OpenHands/software-agent-sdk/pull/2206#pullrequestreview-3857035888 The alternative would be to have async paths for everything (besides the maintenance burden of supporting both, I think there is probably more context that Engel could benefit time)

Was having a conversation about this earlier today. One question that kept coming up is, does the agent support rollback features when you terminate/stop? e.g. if you stop in the middle of a file_edit operation does it cancel the entire operation and brings the file back to its before-state, or is the file stuck in an in-between state

Hmm good question I think the answer is no? LLMs are smart enough nowadays to recover from the in-betweens state

I figured it was a no, but good to know about LLM recovery now a days - that answer is good enough for me 😆