Skip to content

Understand and Create Agent Tools

Enabled tools are described to the model as callable functions. The plugin validates arguments and executes tool calls sequentially during a response.

A tool gives the model a named capability beyond ordinary text generation. SmartSite sends the enabled tool’s description and parameter schema to OpenAI; the model may request a call, and WordPress then validates and executes the saved implementation. Data tools return information, while action tools can send email, call an HTTP endpoint, or invoke WordPress code.

Start with the smallest safe step: select add tool. Do not consider the task finished before you test with safe inputs, inspect the result, then enable and run an end-to-end visitor test; this is where the configuration is tested in the context that truly consumes it.

Use tools only when the assistant needs current structured data or a controlled side effect beyond knowledge retrieval.

Use this feature in the following situations:

  • The assistant needs a small piece of current structured data that is not appropriate for the knowledge base.
  • A visitor request should trigger a controlled contact, webhook, or custom WordPress workflow.
  • You need to limit a capability to one assistant and keep it disabled until testing is complete.
WordPress locationWordPress Dashboard → AI Website Chat → Tools
  • SmartSite Assistant is installed and activated.
  • You are signed in with an account that can manage WordPress options.
  1. Select Add Tool.
  2. Choose Static Data, WordPress Option, Email Action, Webhook Action, or WordPress Hook Action.
  3. Enter a unique normalized name, display name, model description, and visitor progress label.
  4. Choose all assistants or one local assistant ID.
  5. Define parameters and type-specific settings.
  6. Save with Enabled off.
  7. Test with safe inputs, inspect the result, then enable and run an end-to-end visitor test.

A tool gives the AI a controlled way to obtain data or perform an action that ordinary knowledge cannot provide. Its name and description help the model decide when to call it, while the implementation determines what actually happens. Narrow descriptions improve tool selection; overly broad descriptions can cause irrelevant calls, and unsafe implementations remain unsafe no matter how carefully the tool is described.

Fields, controls, and important values
Field, control, or statusWhat SmartSite Assistant does with itHow to use it and why it matters
Name Normalized lowercase function name using letters, numbers, underscores, or hyphens. Evaluate “Name” as part of the complete tool, including selection, arguments, side effects, and returned data. A useful description can improve tool choice, but only defensive code can prevent disclosure or unwanted actions.
Display name / Description Administrator label and model-facing explanation of when the tool is appropriate. “Display name / Description” needs plain language suited to where it appears. Explain the real scope without marketing claims, then read it in context; model-facing descriptions may influence behavior, while visitor-facing copy mainly improves understanding and prompt quality.
Progress label Visitor-facing status text while the tool runs. Write “Progress label” for the person or system that consumes it: a visitor, an administrator, or the model. Specific wording can guide better questions or tool selection; vague promises and mixed purposes make expectations—and sometimes responses—less reliable.
Assistant Blank means all assistants; a selected local ID limits availability. The choice in “Assistant” decides where this feature is available. It may affect who can ask the AI or which profile sees a tool, but it does not make the resulting answer safer unless the underlying operation also checks access.
Enabled Only enabled tools are exposed and executable at runtime. “Enabled” turns the described behavior on or off while leaving its saved configuration in place. Prepare the dependent values first, then compare both states; disabling a feature does not automatically remove logs, remote data, or earlier content.
Success / Error messages Configured feedback returned around execution outcome. “Success / Error messages” can influence when the model calls a tool or what context the call returns. Inspect the entire output—not just the friendly answer—and make the implementation safe before exposing the definition to normal conversations.

Use a separate test session to confirm Understand and Create Agent Tools. This keeps existing login, browser storage, and response history from hiding the change, and it shows whether the result reaches the complete workflow rather than stopping at WordPress storage.

Create a static opening-hours tool so the assistant can return a controlled JSON schedule.

  • Change one part of Understand and Create Agent Tools at a time and keep a short record of the previous value and test result.
  • Use synthetic arguments and inspect the complete raw result before letting the model call the tool in visitor conversations.
Common problems and focused checks
ProblemWhat to check and what to do next
Understand and Create Agent Tools is missing or does not match this guide. Confirm the plugin is active and the account can manage WordPress options. Keep the tool disabled while comparing its schema, arguments, implementation result, and any side effect.
A change on Understand and Create Agent Tools does not produce the expected result. Keep the exact notice and test case, then review the browser console and WordPress/PHP log. Keep the tool disabled while comparing its schema, arguments, implementation result, and any side effect.
Understand and Create Agent Tools
Capture
Show the Add Tool form at the type-selection stage with all five verified tool types visible.
Show
Tool type cards, Name, Display name, Description, Progress label, Assistant, Enabled
Viewport
Desktop, 1440 × 900
Annotate
Use numbered callouts only for controls referenced in the procedure.
Redact
OpenAI keys, tokens, secrets, personal information, private URLs, IP addresses, and conversation text