Skip to main content
A Project’s Guardian is the runtime that evaluates input against Guard Policies. It is exposed:
  • as an API endpoint (POST /v1/guard/{api-id}/analyze) for FOR_API projects, or
  • via the Desktop Agent for FOR_DESKTOP_APP projects.
Open a Project, then click the Guardian card to edit its configuration.
Project detail showing the Guardian card with input types and attached policies

​
Guardian detail

Guardian detail page with Overview, Model Configuration, and Guard Policies sections
The Guardian detail has three sections:

​
Overview

  • Input Types — TEXT, IMAGE, DOCUMENT, and/or ARCHIVE. Requests with a data shape that does not match the bound input types are rejected at the edge.
  • Guardian Engine — the model backend (e.g., vLLM-OCR).

​
Model Configuration

JSON tuning for the analysis model. Typical fields: 
{
  "logprobs": false,
  "top_logprobs": 6,
  "confidence_threshold": 0.7
}
Use the pencil icon in the header to edit. Changes take effect on the next request.

​
Guard Policies

The list of policies attached to this Guardian. Click Add to attach a policy authored in Policies. A Guardian typically carries one PII policy and one Topic policy, but you can mix as needed.

​
Deployment flow

1

Create a Project

In the Organizations workspace, create a Project and choose FOR_API or FOR_DESKTOP_APP. The Project type is fixed once created.
2

Author policies

Create the PII and Topic policies your Guardian will use. See Build a Guard Policy.
3

Configure the Guardian

Pick input types, tune model configuration, and attach policies.
4

Test

For FOR_API projects, issue an API key and call the endpoint. For Desktop App projects, assign end users and confirm traffic is captured.
5

Monitor

Open Opticon from the project header to watch live decisions.

​
Hand-off

Guard API quickstart

Authentication, request/response shape, and error codes for developers calling the endpoint.