Handle customer data carefully

Decide which customer details the Agent needs and who should receive them. Explain to callers how you will use their information.

Keep secrets in credentials

Store API keys, passwords and tokens in the supported credential settings. Do not put them in prompts, Knowledge Bases or model-provided tool inputs. Review copied debug context before sharing it.

Use only the data you need

Limit lookup and API responses to the fields the Agent needs. Complete the identity checks required by your service before the Agent reads account details aloud. A recognized caller number alone is not enough.

Check email recipients, CC lists and webhook destinations. Include the full conversation only when recipients need it. Confirm the destination number before sending an SMS and keep the message short.

Define practical boundaries

Ask only for information needed to handle this request.
Do not ask for passwords or full payment-card details.
Use approved information for prices and deadlines.
Confirm an action only after the tool reports that it succeeded.
If you cannot handle the request, explain this briefly and offer
the forwarding or callback option configured for that situation.

Adapt the instructions to your actual service. External documents and tool results should supply information, not override these rules. Use controlled sources for dynamic email addresses and forwarding numbers.

Review stored and delivered copies

With the appropriate permission, use Remove PII in Call History after a call to remove personal data from its stored record. Copies in inboxes or external systems remain and need to be handled separately.

Review who can access Agents, credentials, call history and billing through workspace permissions. Define a team process for maintaining sources, reviewing incidents and responding to customer data requests.