The agent configuration screen opens when you click a row or the gear icon on the Configure Agents screen. Most fields are disabled for the Frontline agent.

Agent Title

The display name for this agent in the admin console. This is for your reference only and does not affect the agent's behaviour.

Agent Status

Controls who can interact with this agent.

The recommended workflow is to set status to Testing first, verify behaviour in Admin Chat, then switch to Live. The Frontline agent is always live and its status cannot be changed.

Agent Role

A plain-language description of the types of questions this agent handles. The Frontline agent reads this description when deciding whether to delegate an incoming question to this specialist.

Write a clear, concise description of the agent's domain — for example: "Handles questions about pricing, billing, and payment options."

Initial Prompt

The system prompt file used as the starting point for this agent's conversations. Select from the dropdown, which lists all available agent prompt files.

Create and edit prompt files on the Edit Prompts screen.

Agent Prompts

The Agent Prompts section lets you attach one or more prompt files to this agent and configure which functions each prompt can call.

• Click Add Prompt (dropdown) to attach an additional prompt file.
• Expand a prompt card to see its available Functions — tick the ones this agent should be able to use (e.g. web search, product lookup).
• Checked functions are sorted to the top of the list for convenience.
• Click Edit on a prompt card to jump directly to the prompt editor.
• Click Remove to detach the prompt from this agent (the prompt file itself is not deleted).

AirScript Variables

AirScript variables are configuration values that your agent's functions can read and write at runtime. They're used to store settings like API keys, email addresses, and other values that functions need to operate.

What Are They For?

When a function needs external configuration — such as an API endpoint, authentication token, or email address — it reads these values using air.get_config('variable_name'). This keeps sensitive data out of your function code and lets you configure different values for different agents.

Regular vs Secret Variables

Automatic Variable Creation

When you add a function to an agent (by ticking the checkbox in the Agent Prompts section), any variables that function requires are automatically created for you. The system reads the function's code, extracts all air.get_config() calls, and creates corresponding variables.

• Variables with names containing words like token, key, password, or secret are automatically marked as secret.
• Variables with default values in the code are pre-filled with those defaults.
• After saving, you'll see a notification telling you how many variables were created — scroll down to the AirScript Variables section to fill in their values.

Managing Variables Manually

You can also create, edit, and delete variables manually:

• Add Variable — Click the + Add Variable button, enter a name (alphanumerics and underscores only), a value, and tick Secret if the value should be encrypted.
• Edit — Click the pencil icon next to a variable to change its value or secret status.
• Delete — Click the trash icon to remove a variable.

Changes to variables are saved when you click Save Agent Configuration.

Using Variables in Functions

In your AirScript code, read variables with:

api_key = air.get_config("api_key")
email_to = air.get_config("notification_email", "default@example.com")

You can also set variables at runtime (values persist for the conversation):

air.set_config("call_count", current_count + 1)

See AirScript Functions for full details on get_config and set_config.

Search Settings

Title Boost

A score multiplier applied to search results where the query matches the page title. A value of 3 means title matches are ranked 3× higher than body matches. Leave at 1 for no boost.

Search URL Prefixes

Restrict this agent's search results to pages under specific URL prefixes. Only content from URLs starting with one of these prefixes will be returned to this agent.

• Leave empty to allow the agent to search across all indexed content.
• Add one prefix per line, e.g. https://www.example.com/products/.

← Back to Agents overview