Skip to main content
You authenticate once with an organization key, send a request in the OpenAI or Anthropic shape, and get the reply in that shape. The model ID selects the model and the privacy level it runs at.

Three Things Define the API

  • One key, every leading model. A z_sk_ key reaches the whole catalog. You select a model by ID and never hold accounts with the model makers.
  • The privacy level travels with the model. Each catalog entry runs at Anonymous, Private, Confidential, or Encrypted. Frontier models run anonymized. Open-weight models run with zero data retention upstream, and :confidential runs them inside sealed hardware. See Privacy Levels at a Glance.
  • Attestation you can verify. For Confidential calls, Nito gathers the evidence, binds it to a challenge, and returns it through GET /api/v1/attestation. See How Attestation Works.

The Model ID Carries Two Decisions

The base ID names the model. A :confidential suffix asks for the attested variant. There is no privacy field in the request body.
A level never silently weakens. If a model cannot serve the level you asked for, the call is rejected before any provider work.

Principles

  • Nito stores nothing you send or receive. Prompts, completions, reasoning, tool calls, and file contents are request-local. See Data Retention and Visibility.
  • Stronger levels narrow the options. Fewer models, some features off. Match the level to the work by picking the model.
  • Attested, verifiable. Nito brokers the evidence; you verify it.
  • Compatibility first. Most integrations are a base-URL change.

Boundaries

  • Nito does not run your function tools. The model asks, your code runs the function and returns the result.
  • Nito is a gateway, not a model. You always name one.
  • Nito keeps no conversation history.

Next

How the API Works

The request lifecycle.

Quickstart

A working response in minutes.

Privacy Overview

The four levels.