Integrations
Claude Code with Icelake API: Private Coding Agent Setup
Run Claude Code against the Icelake OpenAI-compatible API with a five-minute claude-code-router setup — private inference, usage-based credits, prod or staging base URLs.
What Icelake offers
Fastest way to use Claude Code with Icelake
Install Claude Code and claude-code-router, point the router at https://icelake.io/api/v1/chat/completions with your icl_ API key, then launch with ccr start and ccr code. Full step-by-step lives in the Icelake docs.
Endpoint
POST /api/v1/chat/completions — same OpenAI-compatible path every other client uses.
Router
claude-code-router converts Claude Code traffic to Icelake chat completions locally.
Models
Claude Opus/Sonnet/Fable ids plus coding models from GET /models.
Docs
Copy-paste config and staging URLs on docs.icelake.io/guides/claude-code.
Why Icelake
Why Icelake for Claude Code
Claude Code is a strong coding agent CLI. Icelake gives it a private OpenAI-compatible backend so you control keys, credits, and which models handle agent turns.
Standard chat completions endpoint
No special Claude-only server path. The public Icelake API is the surface Claude Code hits through the router.
Private by default
Icelake does not train on your prompts. Agent sessions stay off mainstream training pipelines.
Prod and staging URLs
Production traffic uses https://icelake.io/api/v1. Point preview deploys at your staging host with the same path shape.
Works with the rest of your stack
The same key and base URL work for OpenAI SDKs, Cursor-style tools, and Open WebUI-style clients.
Setup
Claude Code → Icelake in four steps
This is the shortest path from zero to an agent turn on Icelake. For the full reference (logging, model switch commands, troubleshooting), use the Claude Code docs guide.
- 1
Create an Icelake API key
Sign in at icelake.io → API in the sidebar (or /api/keys). Copy the icl_ secret once. Use a separate key for any staging host.
- 2
Install Claude Code and the router
npm install -g @anthropic-ai/claude-code and @musistudio/claude-code-router@1.0.73 (1.x CLI config.json flow; 3.x is desktop/UI).
- 3
Point the router at Icelake
Write ~/.claude-code-router/config.json with provider name icelake, api_base_url https://icelake.io/api/v1/chat/completions, your API key, Claude model ids (e.g. claude-opus-4-8), and transformer use: ["anthropic"]. Full JSON is in the docs guide.
- 4
Launch
Run ccr start then ccr code (or eval "$(ccr activate)" && claude). Switch models with /model icelake,claude-sonnet-4-6.
Use cases
What developers use Claude Code + Icelake for
Agentic coding without a Claude subscription
Pay for Icelake credits only when the agent runs, instead of locking into a single vendor subscription.
Private refactors and internal codebases
Keep prompts and diffs off training-oriented pipelines while still using a modern coding CLI.
Staging validation before prod
Point claude-code-router at a preview/staging host before production traffic on icelake.io.
Shared team smoke path
Same endpoint shape as OpenAI SDKs and other tools — one API to verify for CI and manual agent tests.
Privacy
Privacy for coding agents
Icelake does not train on prompts or completions sent through the API. Operational logs keep request metadata (ids, model, latency, status) — not your source code payloads.
That matters when Claude Code is reading private repos, env-adjacent files, or customer data during agent runs.
No training on agent prompts
Inference inputs are not used to improve models for other customers.
Key-scoped access
Rotate or revoke API keys from Settings without changing your router install.
Env isolation
Use separate keys for staging and production so agent experiments never touch prod wallets by accident.
Frequently asked questions
Don't see your answer? Contact support or start chatting free.
- Do I need a special Icelake endpoint for Claude Code?
- No. Claude Code uses the standard OpenAI-compatible chat completions URL: https://icelake.io/api/v1/chat/completions. claude-code-router handles protocol translation locally.
- What is the fastest setup?
- Install @anthropic-ai/claude-code and @musistudio/claude-code-router, add a config.json with your icl_ key and Icelake base URL, then run ccr start and ccr code. See docs.icelake.io/guides/claude-code for the full file.
- Which model should I put in the router config?
- Use an id returned by GET /api/v1/models. Claude-family ids include claude-opus-4-8, claude-sonnet-4-6, and claude-fable-5 — always verify the live list.
- How do I test staging vs production?
- Production uses https://icelake.io/api/v1. For a staging/preview host, point api_base_url at that host’s /api/v1/chat/completions and use a key from the same environment.
- Will Claude-specific features work on non-Claude models?
- Claude Code is optimized for Claude. Other coding-capable models on Icelake work for many tasks but may not match Anthropic behavior for extended thinking or Claude-only tooling.
- Where is the full technical guide?
- https://docs.icelake.io/guides/claude-code — install, config.json, staging hosts, smoke curls, and debugging.
Get an API key and open Claude Code
Create a key in Settings, paste it into claude-code-router, and start coding against Icelake in minutes.