Kilo Code v4.82.0…v4.84.1: Modes marketplace, Grok Code, UI improvements
6 releases. 50+ features/fixes.
This week we’ve done 6 releases with over 50 different features/fixes.
Here are the highlights:
New modes marketplace
We now have a curated directory of custom Kilo Code modes.
Why you should care: Different modes are good for different scenarios. Currently we have custom modes like:
Code reviewer
Code simplifier
Documentation specialist
Test engineer
We’ll soon add more modes for different programming languages and more coding tasks.
You can browse the available modes by clicking on the “Marketplace” icon:
Once you “install” a mode, you can easily find it when you open the ‘modes’ dropdown in the chat box:
A new, free model (Grok Code)
We’ve partnered up with xAI to make Grok Code free for 7 days inside Kilo Code.
To use Grok Code, choose the Kilo Code provider (should be the default provider) and choose grok-code-fast-1 from the list:
Automatically remove inference providers with lower context windows
Different inference providers offer different context window sizes for different AI models. This applies mostly to open-source AI models.
Take Qwen3, for example. According to OpenRouter, there are inference providers for Qwen3 that offer it at a 262k context window. Some offer it at a 131k context window.
The Kilo Code provider allows you to choose different inference providers for different AI models. However, not many people use that feature. Most of you want us to choose the best inference providers for you.
We’ve made a small, but significant change in that direction (thanks to @chrarnoldus!)
The change: We now treat OpenRouter as the “source of truth” for the official context size for an AI model. For Qwen3 Coder for example that context size is 262,000 tokens.
Any inference provider whose context window is smaller than the official context size will be ignored.
The goal is to give you the best possible experience when choosing an AI model in our Kilo Code provider.
UI improvements that make Kilo Code easier to use
We shipped 2 UI improvements that will make Kilo Code easier to use:
Improved context menu placement (thanks gerardbalaoro!): We moved the Kilo Code context menu away from the "Navigation" section of the context menu. The context menu now has its own section closer to the Copilot context menu.
The providers list is now searchable and sorted alphabetically (thanks @kevinvandijk!). We keep adding new providers consistently and the current list was long & hard to search. That’s no longer the case:
Add Sonnet 1M context checkbox to Bedrock
https://github.com/Kilo-Org/kilocode/pull/2021
Other updates
We’ve also fixed a few bugs and shipped a few other minor features:
Fixed an issue that caused the same error to be reported multiple times (thanks @chrarnoldus!)
Added a single-file read tool that works better with Sonic than the default multi-file read tool (thanks @daniel-lxs!)
Added API key support to the Ollama provider, enabling usage of Ollama Turbo (thanks @mcowger!)
Fix: The model selector below the chat now shows the correct model list for Qwen Code and some other providers (thanks @chrarnoldus!)
Checkpoints now work when using Morph fast apply (thanks @chrarnoldus!)
We’ve also pulled the most recent changes from Roo Code (Kilo Code is a fork of Roo Code, but we’re also adding our own unique features as you can see from the changes above).
Merges from from Roo Code
Kilo Code is a Roo Code fork. Kilo Code v4.8.3 is now compatible with Roo Code 3.25.20. Some of the most recent changes that we pulled from Roo include:
Model updates:
Added a Sonnet 1M context checkbox to Bedrock
Claude Sonnet 4 context window is now configurable to 1 million tokens in Anthropic provider (thanks @daniel-lxs!)
Added support for OpenAI gpt-5-chat-latest model (#7057 by @PeterDaveHello, PR by @app/roomote)
Excluded GPT-5 models from 20% context window output token cap (thanks @app/roomote!)
Provider support updates:
Added the IO Intelligence Provider support (thanks @ertan2002!)
Added the new Chutes provider models (#6698 by @fstandhartinger, PR by @app/roomote)
Added minimal reasoning support to OpenRouter (thanks @daniel-lxs!)
Workflow updates:
Added an API for resuming tasks by ID (thanks @mrubens!)
Made the enhancement with task history default to true (thanks @liwilliam2021!)
Refactor the task execution system: improve call stack management (thanks @catrielmuller!)
Added support for AGENT.md alongside AGENTS.md (#6912 by @Brendan-Z, PR by @app/roomote)
Fixed: Remove 500-message limit to prevent scrollbar jumping in long conversations
You can see all of the changes pulled from Roo Code 3.25.14 here.
You can see all of the changes pulled from Roo Code 3.25.20 here
Join our Discord
We have a very active Discord server. If you have any questions about Kilo Code, feel free to jump in and ask!