Invitation Downstream

[acoliver]

I like a lot of the work you've been doing to extend and improve gemini-cli. I found it was difficult to work with them and get things merged in at my workrate. Moreover, they were not interested in optimizing out some bad behaviors (nextSpeakerChecker) nor multi-provider and user control features. So I forked it. I've been keeping up with their main and it has been working well. I use llxprt to cherrypick their changes.

I'd love it if you'd consider submitting pull requests to llxprt. We'll want to support things in a multi-platform way but hooks and tools to better support this project could make llxprt more robust and this project support more models! (and faster)

We have some things I think you'll like:

• todo lists like claude code
• tool call controls (no . for whole filesystem)
• model controls (change parameters)
• multi-model
• performance enhancements
• model profile saves (rather than having to pass all that stuff in just llxprt --profile-load myFavModel)

Take a look and let's talk! https://github.com/acoliver/llxprt-code

[bl-ue]

Hi @acoliver! I've been following LLxprt since I discovered it in Piebald-AI/awesome-gemini-cli#1 and I think it has some real potential! Great work!

We're planning soon to implement support for LLxprt as a backend (see #24), and in order to do that optimally I believe we need the provider/model/configuration setting controls you added to LLxprt exposed via the ACP server implementation. We're actually planning on opening a PR soon for just that.

I could also integrate google-gemini/gemini-cli#4401 into LLxprt if that's something you'd be interested in—it's going to be merged, but from the looks of google-gemini/gemini-cli#5221 it might be a while.

Keep up the amazing work! With all the great stuff you've been adding in those 800+ commits I wouldn't be at all surprised if it really took off one of these days...

[acoliver]

That's outstanding. I'm deeply interested in all 3!

If possible, ensure a user can enable/disable the logging as they feel comfortable but a resounding yes.

Ahh 800 commits. Models love to commit...they don't love to run lint or format...so many things get a second commit -- plus the cherrypicking commits.

I'm looking forward to working with you!

[bl-ue]

Alright, I'm starting on it! One question I have is, it seems like you already have something in place for logging, based on LoggingProviderWrapper. Would I replace that with what I implemented in Gemini CLI, or should I add it alongside of your existing logging?

Models love to commit...they don't love to run lint or format...so many things get a second commit

I see. Have you considered using git commit hooks? Sounds like it would prevent that second commit each time.

[acoliver]

I have the hook. It loves to disable it and finds ways around it. I then resort to swearing. My plan is to migrate to a domain and then I'll use a different repo and use collapsed prs.

So I'd suggest extending what I did. I notice it doesnt have the AI turns presently (a bug). The basic idea took Gemini's version but applied it to the providers because in LLxprt doing it how gemini did it wouldn't apply evenly. However, anything clean you come up with is fine. I'd rather not log toooo many times. I'm thinking once this is in place we might migrate /chat save to it too.

Also I have settings that turn it off and on (I feel giving devs control is important).

Thoughts?

[bl-ue]

My PR actually removes /chat entirely and replaces it with automatic recording and a /resume flag (and --resume CLI arg). Will you please take a look at PR #4401's description? It fully implements a fairly complex system from the ground up. I think it may be more substantial than you're thinking. (I also agree about the setting. In #4401 I added one to turn it on and off.)

[acoliver]

I'm good with it. It may be nice to still be able to name them just because sometimes I'm like @#$% this I'll finish later and have 3 things going but the browser sounds good. Would be good if any dialog is theme and responsive (see the /model) but I dig it! (I realize the video it shows the first line that is probably enough, would reocmmend looking at the model dialog and search because if we have 10000 of these it could get awkward without the search but just a recommendation)

…

On Wed, Aug 13, 2025, 11:45 AM bl-ue ***@***.***> wrote: My PR actually removes /chat entirely and replaces it with automatic recording and a /resume flag (and --resume CLI arg). Will you please take a look at PR #4401's description <google-gemini/gemini-cli#4401>? It fully implements a fairly complex system from the ground up. I think it may be more substantial than you're thinking. (I also agree about the setting. In #4401 I added one to turn it on and off.) — Reply to this email directly, view it on GitHub <#25 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAKILKYXFQTHKSK3TXNVZT3NNFPHAVCNFSM6AAAAACDMHAYOGVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTIMBZGM4DSNQ> . You are receiving this because you were mentioned.Message ID: ***@***.*** com>

[acoliver]

check this out https://docs.anthropic.com/en/docs/claude-code/sdk
Claude code can even --resume conversations at the command line and presumably add to them