Autocomplete Requests | Consolidated Discussion

[bturcotte520]

Hi Everyone!

Please leave your feature requests regarding Autocomplete here, to increase visibility and reduce redundancy.

Thanks for being a part of the Kilo Code Community!

[bturcotte520]

From @dzemildzigal:

Hi community!

Some of us do not wish to use codestral as an only option for autocomplete within kilocode.

It is really bad UX to just have that model a possibility for autocomplete.

I want to request a feature, like the main config for KiloCode, to be made available for the autocomplete feature of KiloCode.

That way you can fully BYOK or BYOM into KiloCode.

[bturcotte520]

From @sylwester-liljegren:

Hi!

I have been using Kilo Code and I am very impressed on how much more polished it is compared to e.g. Cline. Really good work!

I have noticed one minor feature that I think you should add to enable more use cases for usage with Kilo Code:

Currently, the Autocomplete feature in Kilo Code seems to be dependent on the user being signed in to Kilo Code web app and then pay for credits to use the Autocomplete feature through a codestral model hosted by Misral AI, if I am not mistaken. While I understand this approach initially, my team does already have e.g. a Codestral deployment in Azure provided through a proxy (OpenAI compatible endpoint) that we would rather like to use in order to apply AI features securely while working on customers' codebases.

I would like to e.g. add a section where one can either check for using either the default API configuration like today based on your credentials or provide an custom API configuration that can be used instead. You could include this within that red rectangle as seen in the below image:

I mean, if you were able to provide the possibility to select an API configuration for generating terminal commands as seen in the below image, which I personally found to be freaking awersome, then I guess it should be possible for you to do something correspondingly for the Autocomplete feature.

Once again, you are doing a great job!

Looking forward to hear your opinions on this!

[bturcotte520]

From @exetroncorp:

Feature Request: Customizable Autocomplete Model Settings

Summary

Add configuration options to customize the model used for autocomplete functionality, allowing users to specify alternative providers beyond the current experimental Kilo code provider.

Problem Description

Currently, the autocomplete feature is marked as "experimental" and requires the Kilo code provider, which presents several limitations:

• Corporate Environment Restrictions: Many corporate environments block or restrict access to external AI services, making the Kilo code provider inaccessible
• Limited Model Options: Users are locked into a single provider/model without flexibility to choose alternatives
• Dependency Issues: Reliance on a single external service creates potential availability and reliability concerns

Proposed Solution

Implement a settings panel or configuration option that allows users to:

1. Select Alternative Providers: Support for self-hosted models such as:

   • Codestral (self-hosted)
   • Qwen (local deployment)
   • Other compatible code completion models

2. Custom Endpoint Configuration: Allow users to specify:

   • Custom API endpoints
   • Authentication credentials (API keys, tokens)
   • Model-specific parameters

3. Provider Fallback Options: Configure primary and fallback providers for reliability

Use Cases

• Enterprise Users: Companies with strict security policies that require self-hosted solutions
• Privacy-Conscious Developers: Users who prefer to keep their code on local/private infrastructure
• Cost Optimization: Organizations wanting to use their own model deployments to control costs
• Specialized Models: Developers working with domain-specific models fine-tuned for their use cases

[Thelm76]

Also in the LiteLLM case, it would be great to add the possibility to chose from available models. My company exposes codestral as vertex_ai/codestral instead of codestral/codestral-latest, so it falls back to claude sonnet 3.7 (see #5861 for reference).

[dzemildzigal]

I would like to bump this up as well:

This feature would make KiloCode completely free and SOTA. A true GitHub Copilot /Cursor/Windsurf rival. Even more than what it is today.

Lets enable BYOK! We would gladly contribute!

[KronisLV]

I've also used Cline and RooCode and BYOK autocomplete would set KiloCode apart from both of those.
There is also Continue.dev but in my experience their autocomplete integration just doesn't work that well and overall the other tools are better than Continue.dev. If KiloCode had that sort of a feature, then it'd become my daily driver over all the others.

Ideally, with the ability to customize the providers/models for autocomplete in the same way we can for the regular functionality.

[antouhou]

Are there any plans to bring the autocomplete to JetBrains? Thank you!

[LigiaZ]

Hi @antouhou, autocomplete is now in JetBrains.
Thank you

[nimrodmoshe05-glitch]

+100000000000 for BYOK Autocomplete!

[paulincai]

Ok, after this number of messages, I think it is time for Kilo team to let us know if LLM selection for Autocomplete is coming any time soon.

[ilyapashuk]

hello
Some time ago you introduced your new auto completion feature.
It is nice, but sometimes I want to disable it, and sometimes - to enable.
My request:
instead of single global switch in the deep settings, add the following options:

1. When it is disabled, allow to temporarily quickly enable it for the current session without changing the global settings, through the status bar for example;
2. global flag in the kilocode settings to fully enable/disable it, remains hear;
3. an other flag, with lower priority then the 2, to also enable/disable it, but implemented in the vscode settings json. This will allow to enable/disable this feature per language, per workspace, per remote...
4. when it is enable, button to temporarily stop it;
5. some scheme, for example special file in the workspace, to block it unconditionally, for example if it is unwanted do to privacy reasons
   your auto completion is a cool feature, but it should be mor flexable.

[luxl]

I'm using Jetbrains kilo plugin v4.134.0, which add Autocomplete recently. But it doesn't support select custom model, but I found a workaround #3827 (comment). I add ["openai":"gemini-2.5-flash-lite"].
But it will not work if your first profile which is "OpenAI Compatible" doest't contain tha model(failing back to be the same as what you choose in Provider menu). I have to delete or edit the first profile to make sure it is "OpenAI Compatible" and contains the model(gemini-2.5-flash-lite). This profile is for Autocomplete. Add other one for chat with different model if you like.
It would be much better if we can choose which profile to use in Autocomplete Menu, or even better, set your custom provider and model. We can add some suggestions to help user selecting the more suitable one, "not thinking-model" for example

[kiolygenius]

It would be great if we could use any model that supports FIM (Fill-In-the-Middle), such as DeepSeek.

[perfectcream]

maybe you can try "google/codegemma-7b" that support Code Completion, include fim_prefix, fim_suffix and fim_middle.

google/codegemma-7b · Hugging Face
https://huggingface.co/google/codegemma-7b

[jcalfee]

Does this really upload every file I work on? Being on by default is a huge security problem. What else might exists like this that I missed?

I have it off now but it was on by default. I would like to use auto-complete, but only on our own hardware / Intranet.

[deliciousbob]

Hi KiloCode Team, we get the most qustions from our users about that autocomplete option.
In Version <= 4.99 the selection of an Autocomplete Model was possible in the GUI, can you pls bring that back to the latest version?
Why dont you give a standard recommended Setting, and allow the users to bring there local models (like described in the docs: https://kilo.ai/docs/basic-usage/autocomplete). Thx alot.
Best regards, Robert

[xolotl18]

what is the state of autocomplete with local openai-compatible endpoints? I cannot suggest the use of this extension at work since we rely on locally hosted models for data security

[Joelius300]

Same, will have to look into alternatives (tabby?) even though Kilo looked promising at first.

[richardchen874-sys]

what is the state of autocomplete with local openai-compatible endpoints? I cannot suggest the use of this extension at work since we rely on locally hosted models for data security

I agree with this. For work environments, autocomplete is often more sensitive than chat because it can run continuously in the editor and may see code before the developer explicitly asks a question.

So local or private OpenAI-compatible endpoints should probably be a first-class autocomplete option, not only a chat/agent option.

I think the autocomplete config should be separate from the main agent/chat config because the requirements are different:

• very low latency
• FIM support
• smaller context window
• workspace / language-level enable-disable rules
• clear privacy boundary
• separate model selection from chat/agent tasks
• optional fallback if the autocomplete model is unavailable
• visible indication of which provider is being used

For teams, this is not only a convenience feature. It affects whether the tool can be approved for customer codebases or internal repositories.

A good design might be:

• default Kilo provider for simple users
• custom OpenAI-compatible endpoint for teams
• local/self-hosted model support
• explicit autocomplete provider profile
• separate budget / rate-limit settings from agent mode

That would make Kilo Code much easier to evaluate in companies that already have internal model gateways or local coding models.

[Fmstrat]

I was thinking of switching us from Continue.dev to Kilo Code, but no way to use local models for auto complete? This is a huge missing feature and deal breaker for us.

[ilukashin]

Same for me.
I am using Continue.dev (FIM autocomplete + small chat) & Zoo Code (agent) which can be replaced by Kilo Code. But there's no support for autocomplete via openai compatible api.