LLM API for Python

Start with the Python client when you want the lightest path to a live request. The LLMWise SDK wraps the same API your dashboard uses, so local testing and production behavior stay aligned.

pip install llmwise

Generate a single `mm_sk_...` key from your LLMWise account and initialize the client once. You can reuse that same key across GPT, Claude, Gemini, DeepSeek, and the rest of the supported catalog.

from llmwise import LLMWise

client = LLMWise("mm_sk_...")

Use `model="auto"` unless you have a reason to pin a specific provider. Auto gives you the cheapest healthy model that fits the request, which is usually the best default for production cost control.

response = client.chat(
    model="auto",
    messages=[{"role": "user", "content": "Summarize this support ticket."}],
)

When you are still choosing the right model for a workflow, call compare mode and inspect the returned model, latency, and cost metadata. This is much faster than wiring multiple provider SDKs manually.

comparison = client.compare(
    model="auto",
    messages=[{"role": "user", "content": "Draft a launch email."}],
)

As traffic grows, keep Auto for default routing and add BYOK only where you need direct provider billing or committed-spend alignment. That lets you stay on one Python surface while changing the billing and routing policy underneath.