Framework integrations

MyIP OpenRouter implements the OpenAI Chat Completions shape. That is why we ship no SDK of our own, and do not need one: anything that can be pointed at an OpenAI endpoint can be pointed at us.

The three settings that matter

Whatever the framework, only three values change.

Our examples use the environment variable MYIP_API_KEY. If a tool only reads OPENAI_API_KEY, put our key there instead.

export OPENAI_BASE_URL="https://openrouter.myip.co.kr/api/v1"
export OPENAI_API_KEY="sk-mo-v1-..."

Frameworks with a dedicated page

• OpenAI SDK — the official Python and TypeScript SDKs. Start here.
• Vercel AI SDK — via @ai-sdk/openai-compatible
• LangChain — point ChatOpenAI at our base URL

Everything else

The tools below have no page of their own but connect with the same three settings. We do not ship or endorse an official integration for them; follow each tool's own documentation.

Check these before you connect

1. Is the endpoint one we implement?

We implement exactly these:

POST /api/v1/chat/completions
POST /api/v1/completions
GET  /api/v1/models
GET  /api/v1/models/{author}/{slug}/endpoints
GET  /api/v1/generation
GET  /api/v1/key       ·  GET /api/v1/auth/key
GET  /api/v1/credits
GET  /api/v1/keys      ·  GET/PATCH/DELETE /api/v1/keys/{hash}   (management key)
GET  /api/v1/providers
GET  /api/v1/datasets/rankings-daily · /api/v1/datasets/app-rankings · /api/v1/benchmarks

A tool that reaches for embeddings, images, audio, the Responses API, or Batch will get 404 not_supported for those features only. Full list: Unsupported endpoints.

2. Does the tool substitute its own model id?

Some tools fill in a default like gpt-4o when you select "OpenAI-compatible". That id is not in our catalogue, so you get 400 model_not_found. Make sure the model name is editable.

3. Does it assume USD?

Every amount we return is in Korean won (KRW). The pricing values from GET /api/v1/models are won per token, pricing.currency is "KRW", and the X-MyIP-Currency: KRW response header says the same. Cost displays written for openrouter.ai may read these numbers as dollars.

4. Are you using the right kind of key?

• sk-mo-v1-… — inference keys, for chat and completion calls.
• sk-mo-mgmt-v1-… — management keys, only for the /keys routes that create and delete keys.

Calling an inference endpoint with a management key returns 401 invalid_api_key.

Smoke test first

Checking with curl before wiring a framework removes half the possible problems.

curl https://openrouter.myip.co.kr/api/v1/chat/completions \
  -H "Authorization: Bearer $MYIP_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/gemma-4-26b-a4b",
    "messages": [{"role": "user", "content": "ping"}]
  }'

Check the key itself:

curl https://openrouter.myip.co.kr/api/v1/key -H "Authorization: Bearer $MYIP_API_KEY"

Attributing your apps

If several apps share one key, tell them apart with the HTTP-Referer and X-Title headers. Most SDKs expose a default-headers option. See App attribution.

When something breaks

The error format and the full table are in Errors and debugging.