Quickstart

Prism is an OpenAI-compatible endpoint. If your code talks to OpenAI today, changing two lines makes it talk to every model in the pool.

1. Get a key

Create a key in the dashboard. Keys are scoped to a project and carry their own budget, rate limit and default policy.

2. Point your client at Prism

from openai import OpenAI

client = OpenAI(
    base_url="https://api.prism.example/v1",
    api_key="pk_live_…",
)

reply = client.chat.completions.create(
    model="prism/auto",
    messages=[{"role": "user", "content": "Summarize this transcript…"}],
)

The model id prism/auto hands the choice to your key's default policy. Any catalog id — anthropic/claude-sonnet-5, say — pins that model and the router steps aside.

3. Read the log

Every response carries an x-prism-route header naming the model that answered and the policy that chose it. The same record, with cost and latency, is in the dashboard within a second.

Next

Write a routing policy, turn on fallbacks, and set a budget on the key before it goes to production.