Tell it what you run. It writes the setup.

Claude Code reads this before it touches the repository.

## AI traffic

Every call to a model provider goes through Lyntway. Set the base URL
to https://lyntway.com/gw/<provider> and present the Lyntway key from
the environment; never write a provider key into code or configuration.
The receipt comes back in the X-Lyntway-Request-Receipt header. Do not
add a direct provider endpoint to work around this.

Applies to every file, so a generated client is built the right way the first time.

---
description: Route model traffic through Lyntway
alwaysApply: true
---

When creating an OpenAI, Anthropic or Gemini client, read the base URL
and key from OPENAI_BASE_URL / OPENAI_API_KEY or ANTHROPIC_BASE_URL /
ANTHROPIC_AUTH_TOKEN. Never hardcode api.openai.com or
api.anthropic.com. Never place a provider key in source.

Manage Language Models → Add Models → Custom Endpoint. The model's url is the full endpoint.

[{ "name": "Lyntway", "vendor": "customendpoint",
   "apiType": "chat-completions",
   "apiKey": "${input:lyntwayKey}",
   "models": [{ "id": "gpt-4o",
     "url": "https://lyntway.com/gw/openai/v1/chat/completions",
     "toolCalling": true }] }]

The token is prompted for and stored, not written into the file.

{ "inputs": [{ "type": "promptString", "id": "lyntwayKey",
               "description": "Lyntway key", "password": true }],
  "servers": { "github": { "type": "http",
    "url": "https://lyntway.com/gw/github/",
    "headers": { "Authorization": "Bearer ${input:lyntwayKey}" } } } }