Pay less per request. Keep your provider.

Prism reads each request your app is about to make and names the cheapest model that can do it. Short questions go to a fast model, hard ones to a strong one. You call the model with your own key, so nothing about your setup changes.

See what you'd save Read the routing docs

Up to 70% off simple requests · $0.0002 per decision · works on any chain · opens at launch

Your app asks PrismSend the chat request you are about to make.
Prism picks the tierFrom what the request needs: length, code, tools, images, reasoning.fast Claude Haiku 5.5mid Grok 4.7top GPT-5.5 Pro
You call it yourselfWith your own OpenAI, Anthropic or OpenRouter key. Prism never sees the key or the reply.

Two lines of integration

Wrap the OpenAI client you already use. Simple requests go to a cheaper model; everything else stays on the model you picked. If Prism is slow or unreachable, the request goes to your usual model, so it cannot break your app.

import OpenAI from "openai";
import { withPrism } from "./prism-route.mjs";

const openai = withPrism(new OpenAI(), { apiKey: process.env.PRISM_ROUTE_KEY });

// use openai exactly as before
const reply = await openai.chat.completions.create({ model: "gpt-5.5", messages });

The drop-in SDK (one file, no dependencies) and the route endpoint ship at launch.

Any provider: Name your own model for each tier: OpenAI, Anthropic, Google, OpenRouter or your own.
Any chain: It is a plain HTTPS API. Solana, any other chain, or no chain at all.
Nothing to migrate: Keep your SDK, your account and your billing with your provider. Add one call.

What would you save?

Most apps send every request to one strong model. Plenty of those requests are simple and do not need it.

Requests a monthModel you use today

Input tokens per requestOutput tokens per request

Requests that are simple: 60%

Simple requests go to Claude Haiku 5.5. The rest stay on your current model.

Estimated saving a month$322

Today: $600
With Prism routing: $278
Routing fees included: $20
Cut: 54%

List prices per million tokens from OpenRouter, read live, plus $0.0002 per routed request. Your real mix will differ.

Built for anything that calls a model

Telegram and Discord bots: Most messages are short. Stop paying top-model prices for a greeting.
Launchpads: Token descriptions, summaries and moderation at a fraction of the cost.
DEXes and perps: Trade explainers and assistants that grow with volume, not with your bill.
Apps and agents: Route every step of an agent to the cheapest model that can do it.

Your setup stays yours

Prism picks the model. That is all it does.

Never sees your keys: You call the model yourself. Prism never touches your provider account.
Never keeps your requests: The request is used only to pick a model and is not stored. The reply never passes through Prism.
Cannot break your app: If Prism is slow or down, the request goes to your usual model. A routing key can only pick models; it cannot spend.

Pay per decision

Routing only$0.0002 per routed request

$0.20 per 1,000 requests, whatever their size
Your provider bills you for the model as usual
No monthly fee, no minimum, no $PRISM needed

Prefer one bill?Prism API

Skip your own provider: call hundreds of models through Prism
Pay with NOTE at each model's price plus a 5% fee
See the developer docs

Route your first request at launch

Read the docs now; the key, the SDK and the endpoint open when Prism launches.

Read the docs Launch app