A unified LLM provider — closed alpha

One API key.
Every model.

Faucet is an OpenAI-compatible gateway in front of 5 providers. Point your existing client at one base URL, pass one key, and switch models by changing a string. There is no Faucet SDK to install.

Request access Read the guide

Faucet is in closed alpha, so accounts are created by hand rather than signed up for. Interested? Email keunwoopark.524@gmail.com.

Point the AI SDK here

Or the OpenAI SDK, or curl. Anything that speaks the OpenAI wire format already speaks to Faucet.

import { createOpenAICompatible } from '@ai-sdk/openai-compatible';
import { streamText } from 'ai';

const faucet = createOpenAICompatible({
  name: 'faucet',
  baseURL: 'https://api.intfaucet.com/v1',
  apiKey: process.env.FAUCET_API_KEY,
  includeUsage: true,
});

const result = streamText({
  model: faucet('anthropic/claude-opus-5'),
  prompt: 'Hello',
});

One API key, every model

One key and one base URL reach every model in the catalog. Switching between them is a string change, not a second integration, a second credential or a second invoice — and there is no Faucet SDK to install, because the endpoint is OpenAI-compatible.

See every model we carry →

Trusted providers

Requests go to Anthropic, Fireworks AI, Google, OpenAI and Voyage AI directly. No resellers and no unnamed intermediaries — the model you asked for is served by the company that built it, at the same weights you would get calling them yourself. Each of them states in its API terms that it does not train on the data sent through the API, and those are the terms your traffic runs under.

Prompts are never stored

Faucet meters requests, not their contents. Your prompts, the completions and every tool call are relayed to the provider and then gone — what a request leaves behind is the model, token counts, latency and cost. Even a provider error is recorded by its status and code, because the text of one quotes the input often enough to count as your data. There is no payload table to breach and no retention window to ask about.

1% fee

You pay what the provider charges plus one percent. Usage is metered per request — cached and reasoning tokens counted separately, streaming included — so the bill is arithmetic you can check rather than a plan you have to fit into.

Want in on the alpha?

Tell us what you are building and we will set up an organization, fund it with alpha credit, and send you a key.

Email keunwoopark.524@gmail.com Read the guide