We're back on Product Hunt — here's how far Kilo has come
A year ago we launched for the first time. Today we're launching again, with a very different product and a lot of community-earned lessons.
A year ago, Kilo Code launched on Product Hunt as an AI coding extension for VS Code. It was a solid start. Today we’re back, and what we’re launching looks nothing like what we shipped that first day.
Over the past year, Kilo has grown into a full ecosystem for agentic development: VS Code and JetBrains extension, CLI, Cloud Agents, Slack, a code reviewer, and more. The IDE extensions alone have crossed 2.3 million installs. At some point the question shifted from “how do we make the chat window better” to “what does the tooling actually need to look like when you’re running multiple agents at once, across your whole organization.”
What we’re launching today is a big step in that direction.
You’re the orchestrator now
Not everyone works the same way. Some developers want AI as a close collaborator - suggesting, editing, explaining, while they stay hands-on with most changes. Others want to hand off whole tasks and review the results. A lot of people are somewhere in between, and where you sit on that spectrum changes day to day, task to task.
The new Kilo for VS Code is built for the full spectrum. It’s where you can read diffs, leave comments, and accept or reject changes line by line when you want to stay close to the code. It’s also where the Agent Manager lives, so when you’re ready to run multiple agents in parallel you can easily do so.
Parallelism runs at every level in the new VS Code extension. Within a single agent, file reads, searches, and terminal commands execute simultaneously instead of one at a time, you feel that difference in speed immediately.
Across agents, the Agent Manager is your control panel for parallel work. Open multiple Kilo tabs, give each a role, and run them simultaneously. For full isolation, create git worktrees, separate branches of your repository where each agent works independently without stepping on the others. One agent builds the new API endpoint, another refactors the auth module, a third writes tests. When they’re done, you review what changed, apply, commit, or open a PR directly from the Agent Manager.
Kilo comes with built-in agent modes: personas like Code, Plan, Debug, and Ask, each optimized for a specific kind of work. Code writes and edits. Plan maps out what needs to happen before touching anything. Debug investigates and traces problems. Ask answers questions without modifying files. You switch between them depending on what the task actually needs. And if none of the built-in modes fit, you can define custom subagents from scratch - give them a specific role, a preferred model, and scoped permissions so each one is tuned for the job your team actually does.
When you want to run the same task across multiple models, model comparison lets you start Claude, GPT, Gemini - or any of the 500+ models Kilo supports - on the same prompt simultaneously and compare the results side by side. Useful for open-ended tasks like refactors or architecture decisions where you want real options before choosing a direction. No lock-in, no forced defaults, use hosted models, BYOK, or local models and pick the best fit every time.
When an agent finishes a task, you need to be able to review the result in detail. Local code review in VS Code shows every change made, file by file. Leave line-level comments on the diff, the same way you would in a pull request by a colleague, and send all your feedback to the agent as structured context in one click. Once you push and open a PR, our cloud agent code reviewer picks it up and gives you a final pass with line-by-line suggestions.
Codebase indexing means your agents have awareness of your full project structure without you manually providing context or burning tokens on large numbers of tool calls. The agent knows your codebase. You don’t have to explain it every time.
Built for how your whole team works
Not everyone on a team builds the same way. Some people live in their editor. Some prefer the terminal. Some prefer to kick off work in Cloud Agents or a Slack thread and do a local review when the output is ready.
Kilo supports all of those workflows. The same core engine runs across VS Code, the CLI, Cloud Agents, and Slack. A session you start in the terminal you can pick up in VS Code. Work kicked off from Slack lands somewhere you can inspect and iterate on locally.
On top of that, teams get fine-grained model controls and deep usage insights, because you shouldn’t have to choose between giving your team access to the best models and having visibility into how they’re being used.
How we got here: a busy month and a lot of help from the community
Early April, we GA’d the new extension to our full install base. It was the biggest architectural change since launch: a shared engine across VS Code and the CLI (based on OpenCode server — an MIT-licensed, open-source foundation for agentic coding), and a complete rebuild of the extension featuring parallel subagents, the Agent Manager, and worktrees. We’d pre-released to 20% of users and incorporated their feedback before flipping everyone over.
The GA release surfaced things at scale our pre-release couldn’t: memory spikes on Windows, rate-limit edge cases, and some users preferring to stay on v5 while we sorted it out. The community brought it all to us fast and in detail, and that’s exactly what we needed.
What followed was three weeks of public iteration. 188 PRs in week two alone. 21 upstream merges. Windows memory: fixed. Rate limiting: fixed. Session stability: substantially better. We committed to weekly updates and kept that commitment. The extension is in the shape it’s in today because the community showed up — bug reports, heap snapshots, open PRs, Discord threads at all hours.
Join us today
We’re running two sessions today to walk through how the new Kilo actually works in practice:
Kilo Show for Coders: workflows from the engineering team, with a deep dive into Agent Manager, worktrees, and the full agentic coding loop.
Kilo Show for Non-Coders: workflows from our growth team, showing how people who don’t write code every day can use Kilo to get real work done.
Kilo Code is free and open source. Always has been, always will be. You can inspect the code, contribute, and run it without a subscription. Buy tokens at provider rates, without markup.
If you’ve been with us since the beginning, thank you - you’ve shaped this more than you know. If you’re new, this is a good time to start.
Is there a reporting mechanism to allow for audit, pinching, and pruning across different tasks sessions within a single interface, wrapped in an MCP?