SupportWhale
← All help articles
Team & organization

Set up and use AI agents

Configure an AI agent, test its answers, assign conversations and manage credits and human takeover.

Last updated · Oct 27, 2026

Open AI agents

Select the correct organization and open Settings → AI agents. Owners and administrators can create agents and manage assignment rules. Teammates can assign conversations to active agents from the inbox.

Normal AI replies and tests use AI credits and require a paid plan. Check your plan and available credits in Billing before you start. Free does not include live AI usage.

Create an agent

  1. Choose New agent and enter a recognizable name.
  2. Choose Fixed model under Routing and select a model from the available list. The Provider filter narrows that list.
  3. Write Instructions explaining what the agent should help with, how it should answer and when it should ask for human help.
  4. Add reference text under Knowledge, such as product facts, support procedures and frequently asked questions.
  5. Set the tone of voice and language. Automatic language follows the customer; English, Dutch and German can also be selected explicitly.
  6. Save the agent as a Draft so you can test it before using it in customer conversations.

With Automatic routing, also choose a model for complex questions. SupportWhale then selects between your regular and complex-question models based on the incoming message. Available models can change; use the choices currently shown in the editor.

The agent uses your saved instructions, reference text and the customer-facing conversation history. Internal notes are excluded. The agent can provide answers but cannot perform account actions such as changing an order or issuing a refund.

Test saved answers

After saving, enter a representative customer question under Test agent. Check the answer against your actual product and procedures, then adjust the instructions or knowledge and save before testing again. The preview uses the saved version of the agent and does not send a message to a customer.

Tests use Live credits by default. The result shows the credits used. Test credits are only available when separate test access has been enabled for your organization; selecting a draft does not make a preview free.

Activate and assign

Set the agent to Active once you are satisfied with its answers. In the inbox, open a conversation and select the agent in the assignment menu. The agent can then answer the latest customer message and subsequent messages while assigned.

To automate assignment, add an Assignment rule on the AI agents page. Choose an active agent and, if needed, conditions for language, tags, verified customer identity or a time window with a time zone. All conditions on a rule must match. Rules are checked from the lowest priority number upward; the first matching rule with an active agent is used. A rule without conditions matches any conversation using automatic assignment.

Use Test rules to check which agent would be selected for the supplied customer details. This checks assignment without sending an AI reply. A conversation uses rules when Use assignment rules is selected in its assignment menu. If no rule matches, the conversation stays with your team.

Take over or pause

Choose Take over in an AI conversation to assign it to yourself, or select another teammate in the assignment menu. This stops further AI replies in that conversation. To return it to AI, select an active agent or Use assignment rules again.

Pausing an agent in Settings removes its current AI assignments. Those conversations remain available for your team. Reactivating the agent does not automatically restore those assignments.

Credits, budgets and interruptions

Included credits renew each billing period, while purchased credits have a separate balance. Usage depends on the model and the amount of text processed; one reply is not necessarily one credit. A request already being processed can still use credits if a teammate takes over before its answer is delivered.

Automatic credit top-ups require your explicit consent, an authorized payment method and a monthly budget in Billing. If another pack would exceed that budget, it is not bought automatically. You can continue using credits already available. The extra-conversation budget is a separate setting.

If there are too few available credits, billing is suspended or an AI request fails, the affected conversation returns to manual handling. A teammate can continue it. Resolve the issue in Billing or review the error, then explicitly assign an agent or enable assignment rules again for that conversation.

See plans, credits and invoices for billing controls and analytics and customer feedback for separate human and AI response metrics.

Need a hand? Contact us →