Argus

Configuration

Choose the model provider — hosted Azure AI Foundry, or a local Ollama daemon — connect it, pick the models scans use, and set how the runtime exploit proof (DAST) is isolated. Saved settings take effect immediately (no redeploy). The Foundry key is encrypted at rest and never shown again once stored; Ollama needs no credential.

Model provider

This server scans with Azure AI Foundry. The local Ollama provider is a workstation feature: it needs a model daemon on the same machine, so a deployment cannot use it.

Hosted. Code excerpts are sent to the Azure endpoint you configure below.

Azure AI Foundry

The hosted provider. Endpoint + key or managed identity.

Foundry resource

Detected from your Azure subscription. Choosing one fills both URLs below.Inference endpointWhere scans send model calls. A project URL is accepted and rewritten to the resource's /models route.Project endpointOptional, for listing deployments. Copy from the Foundry portal's project overview.API keyOr leave blank to use the hosting managed identity.API versionOptional. Blank follows the route, which is normally right: /models wants 2024-05-01-preview, /openai/v1 wants none at all.

Models

Deployment(s) a scan uses by default. Multiple are unioned; the strongest confirms.

No endpoint configured yet, so no deployments have been requested. Pick a resource above, or add deployment names by hand.

No models yet. Add a deployment name below.

Repository access

A GitHub personal access token (PAT) so scans can clone private and enterprise repositories. Public repos need none.

GitHub tokenFine-grained or classic PAT with read access to the target repos. Encrypted at rest; used only at clone time, never written into the scan record.

Runtime exploit proof (DAST)

Boots the target app and attacks it. Only runs once an isolation profile is chosen.

Enable DAST by default on new scansIsolation profile

Strict (gVisor + egress-restricted network) is required in production. Off keeps DAST disabled: the fail-safe.