TokenLab OpenAI-Compatible Migration

Move an existing integration to TokenLab without changing how it works. Change configuration first; change code only where the old provider's model IDs or endpoints have no direct match.

What this skill delivers

• A minimal diff or config block.
• The TokenLab base URL and key variable.
• A model mapping from the live catalog, with each mapped ID confirmed.
• A note for each call that needs a native endpoint instead of chat.
• A smoke test command and a one-line rollback.

Inputs to collect

• Where the provider is configured: environment variables, SDK constructors, config files.
• Every model ID in use, and which endpoint family each one calls (chat, Responses, embeddings, images, audio).
• Provider-specific headers or fields the app sends.

Models

<!-- tokenlab:models:start --> <!-- Generated from the live catalog by scripts/sync_models.py. Do not edit; change preferredModels in catalog.json. -->

Operation	Model	Pick	Use
chat	gpt-5.6-luna	default	Example for Chat Completions

This list refreshes daily from the live catalog. Preferred models come first; otherwise models are ranked by their 24-hour success rate. Before a call, confirm the model with GET /v1/models/{model} and read its price from GET /v1/models/{model}/pricing.

<!-- tokenlab:models:end -->

The model above is only an example for a smoke test. Map each model the app uses with tokenlab-model-picker; never assume a name match.

Migration targets

Client	Base URL	Path	Key
OpenAI-compatible SDK	https://api.tokenlab.sh/v1	/chat/completions	Authorization: Bearer $TOKENLAB_API_KEY
Anthropic SDK	https://api.tokenlab.sh	/v1/messages	key in the SDK
Gemini SDK	https://api.tokenlab.sh	/v1beta/models/{model}:generateContent	key in the SDK

Workflow

1. Find the provider configuration, key variables, and model IDs.

2. Change the base URL and the key variable. Keep the SDK, client shape, and request code as they are.

3. Map each model ID: look it up in GET /v1/models (or through tokenlab-model-picker), then confirm the endpoint in GET /v1/models/{model} (accepted_request_formats for chat; public_operations for media). Report any model with no clear match and let the user choose.

4. Keep native semantics: calls that use Responses, Anthropic Messages, Gemini parts, tools, media, audio, embeddings, rerank, or translation stay on their own endpoints.

5. Run the smoke test:

   curl https://api.tokenlab.sh/v1/chat/completions \
     -H "Authorization: Bearer $TOKENLAB_API_KEY" -H "Content-Type: application/json" \
     -d '{"model": "MODEL_ID", "messages": [{"role": "user", "content": "Reply with OK."}]}'
   

6. Write the rollback line: the old base URL and key variable.

Rules

• Do not rewrite the SDK architecture when a base URL change is enough.
• Do not flatten Anthropic or Gemini payloads into chat unless the user accepts the behavior change.
• Do not change a model silently; an OpenRouter slug or another provider's name maps only after a catalog check.
• Keep only headers that TokenLab documents or the app still needs.
• Never put keys in source files or prompts.
• Apps with separate base URLs for chat, embeddings, and media: migrate one family at a time.

Quality bar

• The diff touches configuration first; every code change has a reason.
• Every mapped model ID exists in the live catalog and declares the endpoint the app calls.
• The smoke test passes on a real response, and the rollback restores the old provider.

Related skills

• tokenlab-model-picker: map and confirm model IDs.
• tokenlab-native-endpoints: calls that need a native protocol.
• tokenlab-api-integration: new code and client setup.