Documentation

Quickstart

Point the official OpenAI SDK at Prism instead of OpenAI, and use any Prism model id in place of a model name — including Gemini models, through the same client.

Python

from openai import OpenAI

client = OpenAI(
    base_url="https://your-prism-domain.com/api/v1",
    api_key="sk-dprism-...",
)

completion = client.chat.completions.create(
    model="openai/gpt-5.4-mini",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(completion.choices[0].message.content)

JavaScript

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://your-prism-domain.com/api/v1",
  apiKey: "sk-dprism-...",
});

const completion = await client.chat.completions.create({
  model: "openai/gpt-5.4-mini",
  messages: [{ role: "user", content: "Hello!" }],
});
console.log(completion.choices[0].message.content);

Swap providers by changing only the model string — openai/gpt-5.4-mini and gemini/gemini-flash-lite-latest both work through the exact same client, same method calls, same response shape.

Authentication

Create an API key from your dashboard and use it as your SDK's api_key— or send it directly as a bearer token if you're calling the HTTP API yourself. Keys are shown once at creation — store them somewhere safe.

Authorization: Bearer sk-dprism-...

Available models

Streaming

Pass stream: true (or stream=Truein Python) to any chat model and the SDK's normal streaming iterator works unchanged — no different code path than talking to OpenAI directly.

stream = client.chat.completions.create(
    model="gemini/gemini-flash-lite-latest",
    messages=[{"role": "user", "content": "Write a haiku about the ocean."}],
    stream=True,
)
for chunk in stream:
    delta = chunk.choices[0].delta.content
    if delta:
        print(delta, end="")

The final chunk carries usage and a billingfield, same as the non-streaming response — that's when the call is actually billed, since token usage for a streamed response is only known once it finishes.

Image generation

POST /api/v1/images/generations, or client.images.generate() in the SDK. Image models are billed in tokens exactly like chat models — see the model table above.

image = client.images.generate(
    model="gemini/gemini-3.1-flash-lite-image",
    prompt="A watercolor painting of a mountain lake at sunrise",
)
# image.data[0].b64_json is the base64-encoded image

Billing

Your account holds a single token balance. Every call — chat or image — is billed based on the actual input/output token counts the provider returns, deducted from your balance. Different models consume tokens at different rates — see the model table above for tokens per 1M input/output tokens on each model.

A call is rejected with 402 if your balance is at or below zero before the call starts. Because the exact usage of a call is only known after it completes, a single large completion can take your balance slightly negative — check your logs for exact token usage per call. Balances are credited manually — contact the Prism team to top up.

Raw HTTP

If you'd rather not use an SDK, the endpoints are plain JSON over HTTPS.

cURL

curl https://your-prism-domain.com/api/v1/chat/completions \
  -H "Authorization: Bearer sk-dprism-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini/gemini-flash-lite-latest",
    "messages": [{ "role": "user", "content": "Hello!" }]
  }'

Response

{
  "id": "chatcmpl-...",
  "object": "chat.completion",
  "created": 1730000000,
  "model": "openai/gpt-5.4-mini",
  "choices": [
    { "index": 0, "message": { "role": "assistant", "content": "..." }, "finish_reason": "stop" }
  ],
  "usage": { "prompt_tokens": 12, "completion_tokens": 34, "total_tokens": 46 },
  "billing": { "cost_usd": 0.000163, "tokens_deducted": 0.021733 }
}

Errors