Kilo Code v4.87.0 - v4.91.2: Image generation, task completion diffs, Autocomplete is now generally available

7 versions. Over 30 new features.

This post covers 7 versions we released over the past week. Let’s start with the most exciting stuff. These features are unique to Kilo Code:

See new changes after a task completes

This was a highly requested feature.

Let’s say you created a new task where you asked Kilo to make some minor modifications to a file.

You wait a bit for Kilo to do its thing and the task is complete.

You’ll now notice a new “See New Changes” button after the task is complete:

After you click on this button, you’ll see a diff showing all the changes that were made during the execution of the task:

This feature is available in both of our:

• VS Code extension (Thanks @chrarnoldus!)

• JetBrains plugin (thanks @catrielmuller) that was released last week

We renamed “Inline Assist” to “Kilo Code Autocomplete” & made it generally available

We have a philosophy: When naming a feature, keep the name as close as possible to the feature’s functionality.

That’s why we renamed “Inline Assist” to “Autocomplete” (thanks @catrielmuller!)

In addition, you no longer have to enable this feature in the Experimental tab.

Autocomplete is now available directly under “Settings” in its own tab:

We’ve come a long way with Autocomplete; it’s now way faster and better than what it was a few weeks ago. Go ahead, try it out and let us know what you think about it in the comments below!

You can generate images with the Kilo Code provider

We now have a new experimental feature where you can use Kilo Code to generate images (thanks @chrarnoldus!)

You can enable this feature if you go to Settings → click on the ‘Experimental’ tab:

After enabling this feature, you can use any model and simply ask it to generate an image.

In the background, Kilo Code will call the “generate image” tool, which will in turn call Gemini 2.5 Flash Image Preview (popularly known as Nano Banana).

Let’s give it a spin and prompt it to generate an image of a person called “Kilo Code”:

Opt out of data collection for any model

Many model providers often provide free access to a certain model in order to train that model on your prompts.

When using the Kilo Code provider/OpenRouter, you can now explicitly opt out of that data collection directly inside the extension (thanks @chrarnoldus!):

This way, you’ll avoid using any models that train on your data.

A few UX improvements

We also have a few UX improvements that make Kilo Code easier to use:

• The VS Code extension will now warn you when you try to paste an image into a prompt for a model that doesn’t support images (thanks hassoncs!)

• Let’s say you have an image attached to your prompt. You then decide to change to a model that doesn’t support images. The extension will now notify you of this (thanks @hassoncs!)

• Our extension now shows an error message (thanks @chrarnoldus!) when a model reaches its maximum output token limit. You should get this: "The maximum number of completion tokens was reached."

Other updates

We also pushed a security patch and a few model updates:

• Disabled the alwaysAllowWriteOutsideWorkspace flag (thanks @hassoncs!), which allowed prompt injection attacks.

• Added support for Kimi K2 0905 to Chutes, Fireworks, Groq and Moonshot providers (thanks @chrarnoldus!)

• Disabled support for vscode.dev due to unavailable Node.js APIs (thanks @hassoncs!)

• Marked GPT-5 as not supporting the ‘temperature’ setting for the LiteLLM provider (thanks @TimAidley!)

• Fixed an HTTP 500 error with OpenAI-compatible providers when no custom temperature is set (thanks @charnoldus!)

• Fixed the maximum output size of Claude Opus 4.1, which was inadvertenly set to 8192 rather than 32k (thanks @charnoldus!)

Changes merged from Roo Code v3.26.4

We merged over 30 changes from Roo Code in this release. Some notable mentions:

• Added Vercel AI Gateway provider integration (thanks @joshualipman123!)

• Updated the list of default allowed commands (thanks @mrubens!)

• Optimized memory usage for image handling in webview (thanks @daniel-lxs!)

• Showed console logging in vitests when the --no-silent flag is set (thanks @hassoncs!)

Join Our Discord to test our JetBrains plugin

We released an alpha version of our JetBrains plugin a week ago. As you saw above, we’re also actively improving it with a few new features.

Our plugin is not (yet) on the JetBrains Marketplace. You can download it by joining our Discord server and going to the #alpha-jetbrains channel.