Kilo-Org
Inline Assist completely unusable - please consider bringing back Continue-style real autocomplete
Description
Hey, First of all, thank you for all the work put into the product! It is awesome!
Recently, I have been tracking all the autocomplete/ghost writer/inline assist PR's, and I would have to say I am really disappointed with the direction that you guys are heading. The first attempts autocomplete was not perfect, you are right, but the current "autocomplete" (inline assist) is completely unusable (even after the latest PR from 2 days ago). As someone who relies on autocomplete on a daily basis, and sees great potential in the autocomplete feature (I would argue that is THE main standing out feature that tells Kilo apart from Roo/Cline and gives Kilo a edge), that you seriously considering going back (but ultimately forward) the route of the Continue fork, improving on it (tweaking configs, opening up possibility of customizing the system prompt, support for FIM Models, etc). It is by far the most natural and seamless autocomplete experience, not coming close to the current inline assist, which is in no way whatsoever a substitute for autocomplete.
Thanks again for all the really hard work put into the product, and hope that we have some good news about the autocomplete feature again.
Hey @sergiobank, thanks for opening this issue!
Yeah, we originally removed the old autocomplete because it was insufficient for showing multiline edits and other advanced features that we wanted to get to (cursor jump, next edit prediction, etc). So we switched to this new inline assist infrastructure. But I do agree that there was value in the type-into-type autocompletions that we had before. I think in the future we'll end up having a hybrid approach where we use pieces of both in order to continue to support FIM models and other local, smaller models.
Could you tell me a little bit more about what makes the inline assist "unusable" in your opinion? Is it too slow? The suggestions aren't good? It doesn't work at all? Any info you have will help us shape the future of the product!
Thanks!
Hey @hassoncs ! Thanks for the clarification, and thank you for the awesome project! I was really disappointed when you guys took down the autocomplete without any notice and didn't provide any alternative. It is really a red flag for the user when a feature that the user came to rely on is deprecated out of nowhere (because of this I think is fit to bring it back, I'm sure many of us users feel that way). Personally I don't think the inline assist, even if it did work, in any alternative for the autocomplete, for me they are different things. I really think there is great value in bringing back the continue autocomplete fork, as it really did work. This way, people can continue (no pun intended) to rely on the autocomplete feature (which I really think is the main differentiator of Kilo). To be even better, I would add an option to customize the system prompt in the autocomplete mode (similar to all other modes that you can already customize the prompt. It can even be in the same pane in the UI, just add like an autocomplete mode in which the prompt would show in the settings and be customizable). That way, it can really be tailored to work also natively with FIM models, utilizing the model specific promt/template for prefix, suffix and middle, without having all models to fallback to the hole filler prompt. (For example, the Qwen3-Coder-30B can be an excelent choice and would work really well using its native FIM template). For me I think it is the sweet spot between out of the box functionality and user customization option.
As for the inline assist, my experience is that I tried really many many configs and models, and none of them worked for me. It didn't give me any relevant suggestions, the UI is a mess (even after the latest update), and as I said before, even if it did work, it is just not a substitute for autocomplete.
Hope to hear from you soon good news about autocomplete, and thanks again for getting feedback and listening from us.
Perfect, this was the feedback we needed to hear! I'm really sorry we removed that feature so suddenly. We really didn't mean to delete features that were useful.
I'm already looking into bringing back the old implementation in a way that works well alongside the new features so expect some progress here soon. We're far from done! :D
Again, thanks for the detailed feedback- it's really helpful!!
Hey!! That is great news @hassoncs !! Looking forward to going back to Kilo from RooCode when autocomplete is back!! Thanks for the news! 🚀
@sergiobank is absolutely correct. A very important reason why I chose kilo code over roo code is that kilo code has an auto-completion feature. Moreover, the current in-line assistant and auto-completion should not be considered the same thing—they can coexist. I hope @hassoncs can add the original function as soon as possible. Thank you guys.
I agree with sergio, and decent autocomplete is the missing piece that prevents our team from switching to Kilo. I'm very happy to hear from @hassoncs that this is being worked on.
I'm tired of paying $20 a month for Cursor just for autocomplete. I really need a solid alternative to closed-source options like Windsurf or Cursor.
100%- Autocomplete is our number #1 priority! Mark and I are actively working on it now and making good progress. We’ll let you know in the next couple weeks when our improved version is ready!
On Thu, Oct 2, 2025 at 1:36 PM Kwan96 @.***> wrote:
kwanLeeFrmVi left a comment (Kilo-Org/kilocode#2218) https://github.com/Kilo-Org/kilocode/issues/2218#issuecomment-3362673576
I'm tired of paying $20 a month for Cursor just for autocomplete. I really need a solid alternative to closed-source options like Windsurf or Cursor.
— Reply to this email directly, view it on GitHub https://github.com/Kilo-Org/kilocode/issues/2218#issuecomment-3362673576, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABG6KHK5H5YE7ZHDXS2VVWD3VV5FXAVCNFSM6AAAAACFIXJEE6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGNRSGY3TGNJXGY . You are receiving this because you were mentioned.Message ID: @.***>
- #3099 doesn't address the functional aspects of this issue since it's not hooked up yet, but we are making progress - we're going to go down @sergiobank's suggestion of better integrating continue-based autocomplete.
