Kilo Code
The open-source agentic engineering platform
Start new thread

Kilo Code for VS Code - Parallel agents, diff reviewer, and multi-model comparisons

by
•
We've completely rebuilt Kilo Code for VS Code, built on OpenCode server. New portable core, parallel tool calls, subagent delegation, inline code review, multi-model comparisons. Get started: kilo.ai/install

Add a comment

Replies

Best

Been using VS Code for most of my systems projects and the idea of running parallel agents on different parts of a codebase at once is something I've been waiting for. Curious how the diff reviewer handles C and low-level code — going to try it on my kernel prefetcher project this week.

Maker

Running parallel agents on different parts of a codebase at once is something I've been waiting for.

awesome! really looking forward to seeing what you're building with

keep us posted, feel free to join the Discord server for additional support:

happy shipping!

Maker

also fun fact: parallel agents was one of the bets in v7 that took the longest to get right.

help us spread the word on X!

Maker

 you can try out a few different models with agent manager on the same C task, and compare them against each other.

Maker

 any AI models you'd recommend for programming in C?

Rebuilt for VS Code sounds handy. Parallel tool calls + that subagent thing has me curious - less sitting around for runs? I bounce between GPT/Claude, so multi-model compare in-editor could save tabs. Gonna try it on a throwaway repo tonight.

Maker

 lfg! you're in a safe place, is the leading open-source AI coding extension for with 1,000,000+ downloads ✌️

Maker

 sound great. did you manage to give it a spin? curious to your thoughts.

Maker

yes, really looking forward to your thoughts! feel free to join the Discord server at or/and open issues on GitHub:

Parallel agents and subagent delegation is a solid foundation — but how does Kilo Code decide which model to route a subtask to in multi-model mode? Is that user-defined or does the agent pick autonomously?

 the agent picks the model based on the agent mode. You can choose to route between frontier, balanced, or free models.

Maker

   curious what are your preferred AI coding models

  Claude and GPT-5 for reasoning-heavy tasks, Cursor for faster implementation. Would love to see Kilo Code route between them automatically based on task type.

Maker

   great feedback!

also your comment highlights something key. some tools want to lock you into one interface, one way of doing things, adding friction to your workflow. is different. the mission is to build an all-in-one, agentic AI platform that supports you wherever you are.

you choose how you work: your interface (a GUI editor, a CLI, or both), your model (500+ models available), your workflow. go give it a spin at

is the leading open-source AI coding extension for , with 1,000,000+ downloads.

Maker

Parallel agents and subagent delegation is a solid foundation

framing this!

Maker

oh and fun fact: parallel agents was one of the bets in this release that took the longest to get right.

let's spread the word on X -

Congratulations!

Maker

thanks for the continuous support! please help us spread on LinkedIn,

Maker

Follow-up question: What should launch next?

Maker

 Thank you Madalina! Are you using any agentic coding tools already?

Maker

   also curious how you like working with AI coding agents? do you prefer to (a) have a look at the context usage (human-in-the-loop) or (b) review the outputs only (agent-first)?

The line level review comments on agent diffs is a really smart UX call. Feels like the missing link between AI wrote this and I actually trust this going to prod.

Maker

 I agree! And it helps to get a local review as a sanity check before shipping to a public repo

Maker

And it helps to get a local review as a sanity check _before_ shipping to a public repo

IMHO That's a key point in this release. Review before approve, diff before commit. The right balance between capability and visibility.

Follow-up question: How do you like to work with AI coding agents?

Maker

 yeah it's very helpful for my daily workflows

Maker

strong +1! FYI you can read the docs for the full setup and options:

Maker

The line level review comments on agent diffs is a really smart UX call.

Thank you! Best one-line review of this v7 launch so far.

Maker

we just - lfg!

💡 Bright idea

I'm hearing more and more about Kilo Code, and I have to say it genuinely makes me happy. At we try to use as many coding agents as possible, and Kilo Code is really great. To be honest, I haven't tested it in VS Code yet (I mostly use the CLI), but I'll be doing that soon.
One important question: how do you manage to limit token consumption? Do you have any specific optimization strategies (compression, compaction, output control...)?

Maker

I'm hearing more and more about Kilo Code, and I have to say it genuinely makes me happy.

 thank you! absolute fan of what you're building at , your words mean the world.

Maker

oh and please keep us posted how your experience goes with ! also curious what we should build/improve/fix cc

Maker

 really happy to hear you like the CLI and are keen to give our VS Code extension a try.

About your question. Check out this link:

We currently have auto-compaction, input cache (for those models that support it), and we’re experimenting with codebase indexing! There's also other things in the pipeline that will ship soon, so stay tuned.

Maker

We currently have auto-compaction, input cache (for those models that support it), and we’re experimenting with codebase indexing! There's also other things in the pipeline that will ship soon, so stay tuned.

 looking forward to it 👀

Congrats on the launch! How do you handle contention and latency?

Maker

 Git worktrees are used to avoid conflicts among parallel agents, and latency largely depends on the speed of the model you're using.

So - if you need faster responses, you can freely switch to a faster model - and for more reasoning, switch to a more reasoning-intensive, frontier model!

Maker

 any AI models you'd recommend in both cases?

Maker

thanks for the support, Arya!

Maker

also please help us spread the word on LinkedIn -

this is great. Going to use it for building ActorDo.
KiloCode works as a good alternative to github copilot

Maker

 thanks, love to hear what you think coming from copilot.

Maker

 looking forward to it! gives you agentic coding across 500+ models - not a single Copilot-branded autocomplete. more about how both compares here:

enjoy!

 most people use 2-3 models, so 500 are anyway useless.
copilot is actually very good, autocomplete is just 0.1% of it's worth.

however I will test kilocode for the whole team.

Maker

   Fair point on the model usage! Yet part of what makes our Agent Manager so cool is how easy it is to experiment and run different models in parallel, testing the same prompt.
Curious how your Kilo test drive goes! Please share feedback here, or in

   that's true. Exactly that's why I wanna test it.

How does Kilo deal with context limits on large codebases with multiple subagents? Do subagents keep separate context windows or share the same one?
Maker

that's the best part: each agent has their own context window ✌️

 when you let them run on separate git work-trees the context windows stay separated

Maker

also curious: how many agents do you run in parallel?

Parallel coding agents with a diff reviewer is a solid upgrade over linear autocomplete. Curious how you handle merge conflicts when two agents edit the same function simultaneously.

Maker

great q - uses git worktrees to prevent conflict. you can merge without issues!

Maker

oh and one more thing: feel free to join the Discord server for help and support -

 as said, you can work on separated worktrees and solve merge conflicts using a git workflow

Maker

worth mentioning parallel work pays off when sessions are independent, i.e. neither one's output depends on the other, and they are unlikely to edit the same files. things like independent features, module-scoped refactors, a feature plus an unrelated bug fix, trying 2–4 approaches to the same problem.

learn more:

First
Previous
1234
Next