Generate fingerprints for analysis

Set up an AI agent for AI Security fingerprints. Use it to understand and audit agent behavior.

This guide uses Claude Code by Anthropic as an example.

Before you begin

Before you begin, make sure you have:

An AI Security workspace: If your deployment of AI Security is on-premises, use your own URL. If you're working with the F5 AI Security SaaS deployment, the URL could be us1.calypsoai.app or eu1.calypsoai.app. Use this base URL in your environment variables.
An Anthropic API key: Required to create a connection to Anthropic.
An AI Security Anthropic connection: You'll create this in the next step.
An AI Security Agentic project: Agentic projects support session IDs, which generate fingerprints. You'll create this in a later step.

Create an Anthropic connection

Log in to the AI Security UI.
Create an Anthropic connection. Select Connections > Models > Anthropic.
Enter a Display name, Model name, and your Anthropic API key.
Select Test and Connect.

The connection is a standard Anthropic connection. No additional configuration is required for Claude Code or other Anthropic-based agents.

Create an Agentic project and copy the project token

Agentic projects support session IDs, which are used to identify fingerprints.

Create a new project. Select AI Guardrails > Projects > Create project. Set the project type to Agentic.
Enter an Agentic project name, such as AI Agent.
Select the model you want to use. In this case, it should be Claude.
Assign yourself as a project administrator, and select Next.
You'll see the Connect an agent window. Select Generate API key. This is your API token for the project.
Copy the API token. You'll use this value in your environment variables.
Select Finish.

Set environment variables

Set the following environment variables to route your agent through AI Security. Replace each placeholder with your own values.

Variable	Value
`CALYPSOAI_URL`	Your AI Security workspace URL, such as `us1.calypsoai.app`
`CALYPSOAI_TOKEN`	Your AI Security project token
`ANTHROPIC_BASE_URL`	Your Anthropic connection URL
`ANTHROPIC_AUTH_TOKEN`	Your Anthropic API token
`ANTHROPIC_CUSTOM_HEADERS`	A header containing the session ID. See the format below

For ANTHROPIC_BASE_URL, append /anthropic/ and your connection name to your workspace URL. For example, if you use the SaaS deployment at us1.calypsoai.app, and connect to claude as a model, use the following URL:

https://us1.calypsoai.app/anthropic/claude

For ANTHROPIC_CUSTOM_HEADERS, use the following format:

ANTHROPIC_CUSTOM_HEADERS=x-cai-metadata-session-id: <SESSION_ID>

AI Security uses the <SESSION_ID> to link the agent's individual requests together. This supports advanced features such as Agentic Fingerprints. The <SESSION_ID> can be any alphanumeric string, but it must be unique, like a UUID.

Run an agent session

In your terminal, start your agent. For example, if you've installed the Claude Code CLI on your system, run claude.
Send a message at the Claude Code prompt. You can even run hi to create a session.
For a richer fingerprint, send a prompt that triggers tool use. For example you might ask "What time is it?". Claude Code may respond with a command like date in the bash shell.

In addition, you can change models. At the Claude Code prompt, run

/model

The Claude Code CLI allows you to switch between Claude models.