Kimi Code API — OpenClaw + Claude Code Integration

Kimi Code (K2.5) is Anthropic Messages API compatible. One config change and your OpenClaw runs Claude Code on Kimi's backend.

Quick Start: Get Your API Key

1. Open Kimi Code Console
2. Create an API Key → format: sk-kimi-...

Setup 1: OpenClaw — Add Kimi as a Provider + Model

Add to your openclaw.json under providers and models:

// In providers:
{
  "id": "kimi",
  "type": "anthropic",          // Kimi speaks Anthropic protocol
  "baseUrl": "https://api.kimi.com/coding",
  "apiKey": "sk-kimi-..."
}

// In models (or agents.defaults.models):
{
  "kimi/kimi-k2.5": {
    "alias": "Kimi K2.5",
    "params": {}
  }
}

Then use it anywhere in OpenClaw:

• Set as agent model: "model": "kimi/kimi-k2.5"
• Switch in chat: /model kimi/kimi-k2.5
• Use as default for a specific agent

Setup 2: Claude Code CLI — Direct

export ANTHROPIC_BASE_URL="https://api.kimi.com/coding"
export ANTHROPIC_API_KEY="sk-kimi-..."

# Interactive
claude

# One-shot
claude --print "Refactor this function to use async/await"

Claude Code auto-appends /v1/messages to the base URL. No other changes needed.

Setup 3: OpenClaw Spawns Claude Code with Kimi

In OpenClaw, spawn a Claude Code (ACP) session that uses Kimi as the backend:

# In your agent config or via sessions_spawn:
sessions_spawn(
    runtime="acp",
    task="Your coding task here",
    env={
        "ANTHROPIC_BASE_URL": "https://api.kimi.com/coding",
        "ANTHROPIC_API_KEY": "sk-kimi-..."
    }
)

Or configure it globally in openclaw.json so every ACP spawn uses Kimi by default.

API Reference

Raw Call Examples

curl

curl -s https://api.kimi.com/coding/v1/messages \
  -H "x-api-key: sk-kimi-..." \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"kimi-k2.5","max_tokens":1024,"messages":[{"role":"user","content":"Hello"}]}'

Python (no dependencies)

import json, urllib.request

req = urllib.request.Request(
    "https://api.kimi.com/coding/v1/messages",
    data=json.dumps({
        "model": "kimi-k2.5",
        "max_tokens": 4096,
        "messages": [{"role": "user", "content": "Hello"}]
    }).encode(),
    headers={
        "Content-Type": "application/json",
        "x-api-key": "sk-kimi-...",
        "anthropic-version": "2023-06-01",
    },
)
with urllib.request.urlopen(req, timeout=120) as resp:
    print(json.loads(resp.read())["content"][0]["text"])

Gotchas

• Model name mismatch: Request sends kimi-k2.5, response returns kimi-for-coding. Don't assert on the response model field.
• Anthropic format only: /v1/messages works. /v1/chat/completions (OpenAI format) returns 404.
• api.moonshot.cn ≠ Kimi Code: That's the general Moonshot API — different product, different auth.
• Timeout: Set ≥120s for complex prompts.
• Provider type: Always "type": "anthropic" in OpenClaw config — Kimi speaks Anthropic, not OpenAI.