OpenHands

Hi! If I could get a review on this one for better support of subscription LMs that’d be appreciated: <https://github.com/OpenHands/software-agent-sdk/pull/4517|github.com/OpenHands/software-agent-sdk/pull/4517>

I believe we have in the codebase already a function `make_llm_completion` doing the same thing, unifying Completions API and Responses API calls. Could we use that?

If yes, then I’d have a second question:
• why didn’t the implementor agent, and the reviewer agent, see it?

Found it: <https://github.com/OpenHands/software-agent-sdk/blob/25cc8e56a4d029f4f879bdfead1cd21c11d6483d/openhands-sdk/openhands/sdk/agent/utils.py#L644|make_llm_completion> - aha it’s in a `utils.py` :thinking_face:

Maybe that’s why, it’s not “at hand” in `llm.py``… We could move it to a class method in `LLM` ?

Thinking that there’s more than the condenser profile in the codebase: also the title generator uses an LLM completion/responses, and maybe new profiles like Oracle profile or (not sure) a cleanup profile. For all of them the same applies, it seems to me, they should be usable for ChatGPT sub, right

ah, sorry, the PR description is wrong, but I already fixed that

Sorry, did your agent push?
It invents a new one: <https://github.com/OpenHands/software-agent-sdk/pull/4517/changes#diff-107251f5114a3fccff331a5e3d9be4aec09473e1a859dd2231da6f988cc76330R2075|llm::complete> 

ahh, I guess it didn’t :face_palm: , sorry

I approved the PR.  (My personal preference is typically to not hold maintainers’ PRs, even if there are nits in discussion, I just mention the nits too. :sweat_smile: So we don’t have to re-review after the fixes, since typically those are small)