Skip to main content
Claude Code is Anthropic’s agentic coding CLI. Point it at Icelake for private, usage-based inference on OpenAI-compatible models. Claude Code talks to Anthropic’s API by default. To use Icelake you need a small local proxy: claude-code-router (ccr). It intercepts Claude Code traffic and forwards chat completions to Icelake.

​
Prerequisites

  • Node.js 18+
  • An Icelake account with API credits
  • An API key from API keys

​
Fastest setup

​
1. Install Claude Code and the router

npm install -g @anthropic-ai/claude-code
# CLI config.json flow (recommended for docs / automation):
npm install -g @musistudio/claude-code-router@1.0.73
Note: CCR 3.x is the desktop/UI product and does not use config.json the same way. For the copy-paste JSON setup below, use 1.0.73 (or any 1.x with ccr start / ccr code).

​
2. Create the router config

mkdir -p ~/.claude-code-router
Create ~/.claude-code-router/config.json (Windows: %USERPROFILE%\.claude-code-router\config.json): 
{
  "LOG": true,
  "LOG_LEVEL": "info",
  "API_TIMEOUT_MS": 600000,
  "HOST": "127.0.0.1",
  "Providers": [
    {
      "name": "icelake",
      "api_base_url": "https://icelake.io/api/v1/chat/completions",
      "api_key": "icl_YOUR_KEY_HERE",
      "models": [
        "claude-opus-4-8",
        "claude-opus-4-8-fast",
        "claude-opus-4-7",
        "claude-opus-4-7-fast",
        "claude-opus-4-6",
        "claude-opus-4-5",
        "claude-sonnet-4-6",
        "claude-sonnet-4-5",
        "claude-sonnet-5",
        "claude-fable-5",
        "zai-org-glm-5-2",
        "qwen3-coder-480b-a35b-instruct-turbo"
      ],
      "transformer": {
        "use": ["anthropic"]
      }
    }
  ],
  "Router": {
    "default": "icelake,claude-opus-4-8",
    "think": "icelake,claude-opus-4-8",
    "background": "icelake,claude-sonnet-4-6",
    "longContext": "icelake,claude-opus-4-8",
    "longContextThreshold": 100000
  }
}
Replace icl_YOUR_KEY_HERE with your key. Confirm model ids with: 
curl https://icelake.io/api/v1/models
Claude-family model ids work best with the anthropic transformer (same pattern as other OpenAI-compatible Claude hosts). For non-Claude coding models only, you can switch the transformer to openrouter and set Router defaults to that model id. If you edit config.json while the router is running, apply changes with ccr restart.

​
3. Launch

ccr start
ccr code
Or: 
eval "$(ccr activate)" && claude

​
Staging / dev endpoint

Point the provider at the host that serves your preview/staging deployment of the platform API (same path shape: /api/v1/chat/completions). Production hosts:
EnvironmentBase (append /chat/completions for the router)
Productionhttps://icelake.io/api/v1
Use a key created in that environment. For a Vercel preview URL, use https://<preview-host>/api/v1/chat/completions. 
{
  "name": "icelake-dev",
  "api_base_url": "https://YOUR_STAGING_HOST/api/v1/chat/completions",
  "api_key": "icl_YOUR_DEV_KEY",
  "models": ["claude-opus-4-8", "claude-sonnet-4-6", "zai-org-glm-5-2"],
  "transformer": {
    "use": ["anthropic"]
  }
}
Then set Router defaults to icelake-dev,claude-opus-4-8.

​
How the endpoint is used

PieceValue
ProtocolOpenAI Chat Completions
PathPOST /api/v1/chat/completions
AuthAuthorization: Bearer icl_…
StreamingSupported (stream: true)
ModelsFrom GET /api/v1/models
The router converts Claude Code’s Anthropic-shaped requests into OpenAI chat completions for Icelake. You do not need a separate Icelake “Claude Code API” — the standard public chat endpoint is the integration surface.

​
Supported Claude-family models (examples)

Ids available on the public catalog (verify with GET /models):
ModelIcelake id
Claude Opus 4.8claude-opus-4-8
Claude Opus 4.8 Fastclaude-opus-4-8-fast
Claude Opus 4.7claude-opus-4-7
Claude Opus 4.6claude-opus-4-6
Claude Opus 4.5claude-opus-4-5
Claude Sonnet 4.6claude-sonnet-4-6
Claude Sonnet 4.5claude-sonnet-4-5
Claude Sonnet 5claude-sonnet-5
Claude Fable 5claude-fable-5
Non-Claude coding models (e.g. zai-org-glm-5-2, qwen3-coder-480b-a35b-instruct-turbo) also work through the same endpoint when listed in your provider config.

​
Switch models inside Claude Code

/model icelake,claude-sonnet-4-6
Use any id listed under your provider’s models array (and available from GET /models).

​
GUI config (optional)

ccr ui
Opens a browser UI for editing providers and routing without hand-editing JSON.

​
Debugging

# Server / HTTP logs
~/.claude-code-router/logs/ccr-*.log

# App / routing log
~/.claude-code-router/claude-code-router.log
Set "LOG_LEVEL": "debug" in config for more detail. Common issues:
SymptomCheck
401API key prefix icl_, not expired/revoked
402Wallet credits on that environment
404 modelModel id from GET /models for that env
TimeoutRaise API_TIMEOUT_MS; long agent turns need headroom

​
Smoke-test the API (without Claude Code)

Before wiring the router, verify the endpoint: 
export ICELAKE_API_KEY='icl_...'

curl -sS https://icelake.io/api/v1/chat/completions \
  -H "Authorization: Bearer $ICELAKE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-6",
    "messages": [{"role": "user", "content": "Reply with pong only"}],
    "stream": false
  }'

​
Notes

  • Claude Code is built around Claude-specific features (e.g. extended thinking). Non-Claude models on Icelake work for many coding tasks but may not match Anthropic behavior 1:1.
  • Prefer a strong coding model from GET /models as your Router default. Claude-family models may require a paid plan; free tiers can use models like zai-org-glm-5-2 or qwen3-coder-480b-a35b-instruct-turbo for smoke tests.
  • Keep keys out of git; use a local config file only.