Say hello to Kimchi

We've been building Kimchi for a while, and today it's ready for you to try. Sign up and get $50 in free credits, no card required.

Kimchi is a complete coding platform, and to all you need to do is describe what you need, approve a plan, and the agent takes it from there. It builds through each milestone on its own, and when it's done, there's a PR waiting for you. Not a question. A PR.

Getting started (it takes about 60 seconds)

Paste this in your terminal:

$ curl -fsSL https://github.com/kimchi-sh/kimchi/install.sh | sh

On a Mac, Homebrew works too:

$ brew install kimchi

For Windows users we've created a guide to get you going. If you encounter any difficulties during the installation, jump in Discord and someone on the team will help you out.

Once it's installed, just run:

$ kimchi

If you're already using Claude Code or OpenCode, Kimchi will spot your existing setup - MCP servers, skills, config - and ask if you want to bring it over. One prompt, and you're not starting from scratch.

Choose how you want to work

There are two modes, and both are useful depending on what you're building.

/ferment - for when you want to hand it off

Describe the task. Approve the plan. Walk away. Kimchi handles the rest - building through each milestone autonomously until there's a finished PR. It's designed for the kind of work you want done while you're in a meeting or asleep.

/ferment implement a cart abandonment email flow

Coding session - for when you want to stay in the loop

The interactive mode. Chat with the agent, steer it as it goes, and keep a closer eye on what it's doing. Same model orchestration underneath - you just have more control over the direction.

What's running under the hood

Kimchi uses open-source models only. No rate limits, no token caps, and no lock-in to a single provider. It routes between models automatically depending on what the task needs: a reasoning model plans, faster models execute in parallel, and an evaluator checks every milestone before things move forward.

Model                    Role        Context       Output price
kimi-k2.6                Reasoning   262k tokens   $4.50 / 1M tokens
minimax-m2.7             Execution   197k tokens   $1.20 / 1M tokens
nemotron-3-super-fp4     Execution   1M tokens     $0.75 / 1M tokens
Avg. proprietary         -           -             $12.10 / 1M tokens

For context: the average proprietary model costs $12.10/1M output tokens. Kimchi's cheapest execution model is $0.75. That's a 94% difference. You can lock to a single model if you prefer - but the default routing makes a pretty strong case for itself.

Your team's spend, actually under control

Set monthly limits per API key, per user, or per team. You'll get a heads-up before anyone hits their limit - not a surprise invoice after. It's especially useful when you've got CI pipelines, contractors, and engineers all running tasks under the same account.

Come hang out

Join our Discord to share feedback and meet the people building it. We're in there every day.

Let's build something amazing. Kimchi.dev