Getting started

Introduction

What Navige does to a call, and what it never does.

Navige sits between your agents and the models and tools they call. Every call that goes through it is priced, recorded and attributed to the job it belongs to — and, once you are on a paid plan, checked against the policies and budgets you have set before it is allowed to run.

It is a proxy, not a framework. There is no SDK to adopt and no agent to rewrite. In most cases you change a base URL.

What it will not do
Navige does not read your prompts back to you, does not train on your traffic, and does not silently alter a request. When it changes what your agent asked for — a cheaper model, a cached response — that is a lever you turned on, and the console shows every call it fired on.

Start here

The three things to understand

Most of what surprises people about Navige comes down to three ideas. They are short, and they are worth reading before you wire anything up.

Ways in

Four, depending on what you are connecting. They compose: an agent can call models through the proxy and ask /enforce before it does something irreversible.

  • LLM proxy — point your OpenAI, Anthropic, Gemini or xAI client at Navige and every model call is priced and recorded.
  • Decision API — POST /enforce before a tool runs, and act on the answer.
  • MCP — put Navige in front of an MCP server so each of its tools is governed.
  • Usage import — send spend Navige never saw, so the total is the real total.

The base URL

Everything in these docs is relative to one host.

text
https://api.navige.ai

The console you read the results in is at app.navige.ai, and these docs are part of it.

Something here wrong or missing? Tell us — a person reads it.