Get Started

Step 1 — Create an API key

1. Log in to the Kimchi console.
2. Go to Kimchi → API Keys.
3. Click Create API key. Copy the key and store it securely — you won't see it again.

Step 2 — Install the Kimchi CLI

The CLI detects your installed AI coding tools and writes the correct config files for you — base URL, model list, API key, everything.

curl -fsSL https://github.com/getkimchi/kimchi/releases/latest/download/install.sh | bash

This downloads the binary and launches the interactive setup wizard. It will:

• Ask for your API key and validate it
• Auto-detect installed tools (OpenCode, Cline, Zed, …)
• Let you choose global or project-specific scope
• Optionally install GSD agents for multi-agent orchestration
• Write config files — done

Supported tools

See the full CLI reference for manual install options, debug flags, and shell completions.

Step 3 — Make your first request

After the CLI finishes, open your tool and start prompting. Or verify the connection with a raw cURL call:

curl https://llm.kimchi.dev/openai/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json' \
  -H 'Authorization: Bearer $CASTAI_API_KEY' \
  -X POST -d '{
    "model": "glm-5-fp8",
    "messages": [
      {
        "role": "user",
        "content": "Explain what a Kubernetes DaemonSet does in one sentence."
      }
    ]
  }'

The response follows the standard OpenAI chat completions format. Any OpenAI-compatible client works — just change the base URL to https://llm.kimchi.dev/openai/v1 and use your Kimchi API key.

IDE setup recipes

The CLI handles setup automatically, but if you prefer manual configuration or need tool-specific tweaks, follow these step-by-step guides:

Available models

A common pattern is to pair a reasoning model (GLM 5) for planning with a fast model (MiniMax M2.5) for code execution. The CLI and IDE recipes above set this up automatically.

For current pricing and the complete list, see the Serverless models page in the Kimchi console.

Optional — GSD multi-agent setup

Get Shit Done (GSD) orchestrates multiple AI agents — planner, researcher, executor, verifier — each using the model best suited to the task. The Kimchi CLI can install GSD for you during setup.

If you skipped GSD during CLI setup, install it manually:

For OpenCode (GSD 1.x):

npx gsd-opencode

For GSD 2.0 (standalone TUI):

gsd config

See the OpenCode recipe for full GSD configuration examples with model assignments per agent role.

How it works

Your IDE / CLI ──► Kimchi config ──► https://llm.kimchi.dev/openai/v1 ──► Open-Source Model
                                              │
                                     OpenAI-compatible API

• Serverless Model APIs — Kimchi runs models on its own GPU infrastructure. You pay per token. No cluster needed.
• Self-hosted deployments — Run the same models on your own Kubernetes cluster when per-token costs exceed compute costs or compliance requires it. Learn more →

Pricing

Pay-per-token. Separate rates for input and output tokens. The free tier has generous limits — no credit card required.

When you need more throughput, upgrade to a paid plan. See rate limits for details.

Next steps