One base URL, one key, and a model table that says what is available.

01 — Quickstart

The change you make is a base URL and a key.

Over HTTP

curl https://api.garrisonhealthsystems.com/v1/chat/completions \
  -H "Authorization: Bearer $GARRISON_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
        "model": "meta/llama",
        "messages": [{"role": "user", "content": "Summarise this visit note."}]
      }'

From a client you already have

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.garrisonhealthsystems.com/v1",   # instead of the provider's
    api_key=os.environ["GARRISON_API_KEY"],            # instead of the provider's
)

reply = client.chat.completions.create(
    model="meta/llama",
    messages=[{"role": "user", "content": "Summarise this visit note."}],
)

Switching providers is a change to the model string. The agreement behind the new string is already in place.

02 — Models

Every provider, named as provider/model.

Available means you can route to it today under your Garrison BAA. Coming means it is on the way and the form at the bottom is how you tell us which ones matter to you first. We never route patient data to a preview or beta model.

Some newer models at some providers require the provider to keep prompts for a period, which rules them out of the zero-retention plan. The security page says which, and the model list in your account marks them.

03 — Requests

What a request carries, and what we keep.

• Authentication. A bearer token per key. Keys belong to a team; a team has a budget.
• Format. Chat completions, embeddings and streaming, in the request and response shape the OpenAI client library sends. Errors come back in the same shape.
• Log line. Caller key, team, model, provider, input and output tokens, cost at the provider's list price, time, outcome. Kept six years. No prompt text in the log on either plan.
• Retention. On the standard plan, identifiers on the HIPAA Safe Harbor list are removed from prompts and responses before anything is kept. On zero retention, nothing is kept.

04 — Pricing

List price on tokens. Two plans.

You pay what the provider publishes for every token, on both plans. The plans differ in retention. The pricing page has the full comparison.