# Proxy Architecture

Cencurity Engine is a local guardrail server for AI coding tools.

It sits between your IDE and the model provider.

## Core idea

When the model sends code back, Cencurity Engine checks that output right away and can:

- pass safe output through
- hide sensitive output with `redact`
- stop dangerous output with `block`

## Start-here mental model

If you only remember one thing, remember this:

**keep your API key in your IDE, run Cencurity Engine locally, and point your IDE to `http://localhost:8080`.**

Most users do not need to copy their provider API key into Cencurity Engine.

## Simple structure

```text
IDE or agent
  -> Cencurity Engine
  -> model provider
```

## Community and Engine

- Community exposes the same proxy pattern through `http://localhost:18082`
- the standalone engine runs locally on `http://localhost:8080`