Kilo Code
OSS AI coding assistant for planning, building & fixing code
5.0•42 reviews•4.4K followers
OSS AI coding assistant for planning, building & fixing code
5.0•42 reviews•4.4K followers
Build, ship, and iterate faster with the most popular open-source AI coding agent.
2M+ Kilo Coders. 30T+ tokens processed.
Follow us: kilo.ai/social
This is the 4th launch from Kilo Code. View more
Kilo Code v7 for VS Code
Launching today
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
Launch Team / Built With
Augment Code
Cline
VSCoder Copilot
Looking forward to seeing what you're building with @Kilo Code!
Kilo Code
@fmerian thank you, what's your favorite new feature in this release?
@jobrietbergen multi-modal comparisons are so helpful. there's an always-on debate on what the best AI models for coding are. [1] now we know.
[1]: What's the best AI model for coding?
Kilo Code
@fmerian yeah that's very helpful to benchmark performance across different tasks
KiloClaw
Hey Product Hunt 👋 Brian from Kilo Code here.
We just shipped the biggest update to our VS Code extension since launch. The entire thing has been rebuilt on OpenCode server, which is the same open-source core that powers our CLI and Cloud Agents. One engine across every surface, so when we improve something, it gets better everywhere.
The headline feature is real parallelism. Kilo can now run multiple tool calls at the same time (file reads, searches, terminal commands all firing concurrently), and it can spin up parallel subagents that each handle a piece of a larger task simultaneously. You actually feel the speed difference.
A few other things shipping in this release:
Agent Manager — run multiple independent agents in separate tabs, give each one a role, and use git worktrees so they never step on each other's code
Inline code review — leave line-level comments directly on agent diffs, just like a PR review, and send them back as structured context
Multi-model comparisons — run the same prompt through different models side by side and pick the best result
Cross-platform sessions — start in the CLI, pick up in VS Code, share with a teammate
Kilo is open source, runs 500+ models at provider cost (zero markup), and has over a million developers using it. We'd love for you to try it out and tell us what you think!
Kilo Code
@brian_turcotte agent manager is such a great addition, and overall it works so much faster! Great to be working together on this one!
@brian_turcotte @jobrietbergen great product. great team. period.
Kilo Code
@brian_turcotte @fmerian 🫶
@brian_turcotte curious what's your @VS Code look like? any extensions you'd recommend in addition to @Kilo Code?
KiloClaw
@fmerian I added a front-end specialist skill from skills.mp, but other than that - the vanilla extension is all I need for my work!
ClawSecure
@brian_turcotte One engine across CLI, VS Code, and Cloud Agents is the right architecture. We run agent workflows across multiple surfaces internally and the biggest friction is always inconsistency between environments. Something works in the terminal, behaves differently in the editor, breaks entirely in cloud. A unified core that improves everywhere simultaneously solves that at the infrastructure level.
Real parallelism is the feature that matters most here. Sequential tool calls are the hidden bottleneck in most agentic coding workflows. When a complex task requires reading 10 files, searching a codebase, and running terminal commands, doing that one at a time turns a 30-second task into a 5-minute wait. Parallel subagents handling pieces of a larger task simultaneously is where the speed compounds even further, especially on architectural work that naturally decomposes into independent subtasks.
The Agent Manager with git worktrees is a smart detail. We've run into the exact problem of multiple agents stepping on each other's code. Isolating each agent in its own worktree so they can operate independently without merge conflicts is the kind of practical engineering decision that shows you've actually dealt with multi-agent workflows in production, not just theorized about them.
Inline code review on agent diffs is great too. The gap between "agent generates code" and "human approves code" is usually copy-paste into a PR tool. Doing it directly in the editor with structured context going back to the agent tightens that feedback loop significantly.
Open source at provider cost with zero markup across 500+ models. Hard to argue with that. Congrats on the release.
KiloClaw
@jdsalbego Thank you very much for the thoughtful comment!
@brian_turcotte Hey Brian, what stops two subagents from making contradictory edits to the same file when they're working on related parts of a task? You mentioned git worktrees for the Agent Manager; is the same isolation used for subagents, or is there a different mechanism?
@brian_turcotte congratulations on the release! I am new to Kilo Code but am looking forward to trying it out. I commented under@kzhao post because I am also curious about this, and would you still "recommend" agents work on separate tasks or even separate files when possible to avoid conflicts or is the Agent Manager able to handle?
KiloClaw
@kzhao Exactly - it's git worktrees for Agent Manager, and with subagents it depends - if the tool calls won't conflict or aren't actually writing code (such as read or plan agents), worktrees aren't required.
Kilo Code
@kzhao @brian_turcotte @blaize_olle I'd just add that subagents are delegated by a parent agent mainly to keep its context focused - they go do the work, and return a summary to the parent rather than the full tool-call history. They share the parent's workspace rather than getting their own worktree, so the safeguard against contradictory edits there is the parent orchestrating: subagents are best used for scoped exploration and well-bounded changes, while the parent integrates.
Where Agent Manager really shines is when the tasks are already structurally independent (e.g., different features, different files, or "throw the same prompt at three models and pick the best result") - it gives you one surface to track all of them, see PR/review status, and prune what you don't want.
More on that in our livestream that aired earlier today and is now available on demand. Marius, the engineer behind Agent Manager, breaks it down here (jump to 18:00) https://app.livestorm.co/kilocode/kilo-show-for-coders
Product Hunt
KiloClaw
@curiouskitty Good question!
We offer both local reviews in the extension and automated reviews on Github and GitLab PRs.
You can specify the strictness in both cases, so that you can filter down to the review points you actually care about.
Kilo Code
@anusuya_bhuyan they run simultaneously!
framing this!
Kilo Code
@anusuya_bhuyan it runs them simultaneously on separate worktrees
KiloClaw
@anusuya_bhuyan Either!
You can run them in parallel inside the Agent Manager or just start a new branch and run them sequentially.
Does the agent manager have any limits on how many parallel agents you can run at once, or is it just constrained by your machine's resources?
Kilo Code
@barnaby_lloyd it is only constrained by the hardware you’re running on.
@realolearycrew for example, what's your setup and how many agents do you run in parallel?
Kilo Code
@barnaby_lloyd It's constrained by machine resources
KiloClaw
@barnaby_lloyd No limit, so it's a machine constraint if any!
definitely! read this full breakdown: kilo.ai/kilo-code/vs/claude-code
TL,DR: @Kilo Code is open-source, offers multi-model CLI + IDE agent with inline autocomplete vs Anthropic's Claude-only terminal-first coding agent.
eager to have your feedback!
Kilo Code
@lakshminath_dondeti i might be biased, but you definitely should ;) - you can also use your codex subscription inside of Kilo, or use any of the free/frontier models
@jobrietbergen oss ftw!
KiloClaw
@lakshminath_dondeti I certainly think so!
In addition to freely switching models, we're also completely open-source and focused on frictionless agentic orchestration. The openness combined with parallel execution makes it a different experience than other tools, in my humble opinion!
Been on Cursor for a while but the model lock-in is starting to bother me. The 500+ models angle here is hard to ignore. Does switching models mid-project break any context?
Kilo Code
@alexis_rodriguez7 happy to hear, and no, you can even switch model mid task.
exactly - 500+ models, zero markup, no editor switch required. see full comparison vs Cursor here: kilo.ai/kilo-code/vs/cursor#comparison
hope this helps!
KiloClaw
@alexis_rodriguez7 It does not! You can switch in between prompts, or even run parallel agents using different models to see how they handle the tasks differently.
Your context remains intact!