Databricks Primer

1. Provision the workspace

Spin up a Premium-tier workspace in the closest region to your users. Enable Unity Catalog so secrets land in a governed scope, and attach a single-node cluster with Databricks Runtime 14.3 LTS for the warmup calls.

• Premium tier or higher (token-based REST auth)
• Unity Catalog enabled for secret governance
• DBR 14.3 LTS, single-node, 16GB driver

2. Store the Meridian gateway key

Drop your Meridian key into a Databricks secret scope. Never paste it inline in a notebook cell — downstream cluster logs would capture it. Use the CLI:

databricks secrets create-scope meridian
databricks secrets put-secret \
  --scope meridian \
  --key gateway_api_key

# Read it back inside a notebook
import os
key = dbutils.secrets.get("meridian", "gateway_api_key")
os.environ["MERIDIAN_API_KEY"] = key

3. First call from a notebook

With the secret loaded, point the OpenAI-compatible client at the Meridian base URL. Use the adaptive router so the gateway picks the best backend per prompt — you keep one model name across cells.

from openai import OpenAI

client = OpenAI(
    base_url="https://llm.getnimbus.net/v1",
    api_key=os.environ["MERIDIAN_API_KEY"],
)

resp = client.chat.completions.create(
    model="meridian/model-router",
    messages=[
        {"role": "user", "content": "Summarize Delta Lake in 1 line."}
    ],
)
print(resp.choices[0].message.content)