Goal: connect my existing codebase to Kumo — an AI gateway that is drop-in compatible with the major AI SDKs — so my current code keeps working but every model call routes through Kumo. Connection details: - Base URL / endpoint: https://api.kumo.cloud/v1 - Auth header: Authorization: Bearer <KUMO_API_KEY> ^ replace <KUMO_API_KEY> with my key from the Kumo dashboard -> Keys page. - Compatible APIs: OpenAI Chat Completions (/v1/chat/completions) and Responses (/v1/responses), and the Anthropic Messages API (/v1/messages). Keep whichever one my code already uses. - Models: keep whatever model my code already calls. Kumo serves models from several providers (OpenAI, Anthropic, Google and others); the model name stays in the request, exactly as now. Rules: - First detect what my code actually uses (which SDK / HTTP client and which endpoint) before changing anything. - Do NOT rewrite my application logic. Change only the client config: the base URL / endpoint and the API key. - Keep my current SDK / library as-is — Kumo is compatible with it; do not switch me to a different SDK. - Read the key from an environment variable (e.g. KUMO_API_KEY); never hard-code it in source. - Do NOT change my model names. If you can't tell which model or SDK I use, ask me before guessing. Steps: 1. Find where my AI client is configured — the base URL / endpoint and the API key — whatever SDK or HTTP client that is. 2. Point the base URL at https://api.kumo.cloud/v1 and set the key to my Kumo key from the env var. 3. Leave every request, prompt, model name and parameter unchanged. 4. Make one small test call (with a model and SDK I already use) and show me the response to confirm it works. Done when: my code runs against Kumo with only the base URL and key changed, and the test call returns a valid response. Reply: tell me which file(s) you changed and paste the test response.