Skip to main content

AI Model

Connect a model repository or an inference endpoint to simulate attacks against the model and find vulnerable responses (AI attack simulation), and to organize the model's configuration into a bill of materials (AI BOM). Only models you own or have operating authority over are eligible.

  • Analyses : AI attack simulation · AI BOM
  • Connection : Model repository (Hugging Face · GitHub · GitLab) or a model server for evaluation (OpenAI-compatible inference endpoint)
  • What you need : A model repository URL or an inference endpoint URL, depending on the analysis
  • Target : Registered models only. You cannot test an arbitrary model URL entered by hand.
The two analyses connect to different targets

AI BOM reads a model repository to organize its configuration, while AI attack simulation sends requests to a running inference endpoint. The two connections are independent, and you can register just one of them.

What you need​

The connection target depends on the analysis you choose.

  • AI BOM : Connect a model repository. You pick the repository type (Hugging Face · GitHub · GitLab) and enter the URL; private or gated models require a read-access token.
  • AI attack simulation : Register a model server for evaluation. You need the URL of an OpenAI-compatible inference endpoint (https://..., POST /chat/completions) and the model name, plus an API key if authentication is required.
Tokens and keys are stored masked

Repository access tokens and model server API keys are stored encrypted and shown masked on screen. Values are never exposed in plaintext, and leaving a field empty keeps the existing value.

Registration steps​

On the [Services] page, click Register service and select the AI model type to start the four-step wizard.

  1. Service type : Select AI model as the type to register.
  2. Analyses : Name the service and select at least one of AI attack simulation · AI BOM.
  3. Model repository : Enter the connection details for the analyses you selected in step 2.
  4. Confirm : Review the entries and finish registration.

Step 3 connection details​

The input screen differs by the analyses chosen in step 2. If you select both, both areas appear together.

  • Model repository connection (when AI BOM is selected) : Choose the repository type and enter the URL.
    • Hugging Face : Enter the model ID or URL and confirm with Connection test. Leave the access token empty for public models; private or gated models require a read-access token. If you leave revision (branch · tag · commit) empty, main is used, and it cannot be changed after registration.
    • GitHub · GitLab : Enter the repository URL and access token, pass the authentication test, then select the branch to analyze. This is the same connection method as source code registration.
  • Model server for evaluation (when AI attack simulation is selected, required) : Register the model server URL (base_url) and model name to test. This registration is independent of the repository connection, so you can register or change it later in service settings.

When registration finishes, you return to the [Services] list. Open the service and click Run analysis on each analysis tab to start.

AI BOM for GitHub · GitLab models

The AI BOM for GitHub · GitLab repository models is built from a self-authored model specification written by the producing organization, not from public model information. After registration, you record the specification on the AI BOM screen, and when you publish it, the repository's commit and model-file fingerprints are read as well. See AI BOM for details.

Only models you own​

AI attack simulation sends attack prompts to a real model, and AI BOM reads a repository. Both analyses target registered models only, and the target is taken from the registration details, so you cannot enter an arbitrary URL on the analysis screen. Use them only on models you own or have operating authority over.

Settings (changing after registration)​

After registration, you can change the repository access token or the model server for evaluation under [Settings] in the service sidebar. The repository type · URL · revision are fixed at registration. See AI model settings for details.