Using AI Actions with AI Voice Agents

AI Actions allow your AI Voice Agent to take automated actions during live calls. Most come from an integration you connect to Aircall, such as Shopify, HubSpot, Salesforce and Zendesk; a few are native Aircall actions that need no integration at all. Once enabled on an integration and assigned to an agent, the agent can autonomously look up or update data in those tools without human intervention.

This article explains what AI Actions are, which actions are available, and how to configure them. Most follow a two-step model: enabling capabilities at the integration level, then assigning them to individual agents. Native actions, like Send SMS, only need the second step.

Important: Integration-powered AI Actions with AI Voice Agents are currently only available for Shopify, HubSpot, Salesforce, and Zendesk integrations. Salesforce AI actions are available for inbound calls only. Native Aircall actions, such as Send SMS, don't require an integration and follow a different setup path, described below.

What are AI Actions?

AI Actions are capabilities that your AI Voice Agent can invoke during a conversation based on the caller's intent. Most AI Actions come from an integration you connect to Aircall. A few are native Aircall actions that need no integration at all.

Examples include looking up an order status, retrieving a contact record, booking a meeting, or opening a support ticket, all without transferring the call or involving an agent.

Typical examples:

• Looking up an order status from a Shopify store, then telling the caller when it will arrive.
• Identifying a returning caller by phone number and pulling their HubSpot contact record.
• Creating a Zendesk ticket on behalf of the caller.
• Checking calendar availability and booking a meeting in HubSpot for a follow-up call.
• Listing the caller’s open Zendesk tickets so the agent can pick the right one to discuss.

Integration-powered capabilities must be enabled in two places before they work: first at the integration level, then on the specific AI Voice Agent you want to use it. Native Aircall actions, such as Send SMS, only need the second step, since there's no integration-level toggle for them.

Available capabilities by integration

Shopify

Find more information in the dedicated article "AI Actions: Setting up your AIVA to handle Shopify order status calls". 

HubSpot

Note: For integrations that require OAuth authentication, such as HubSpot, a banner will appear prompting you to reconnect before capabilities can be enabled. The capabilities sync toggle activates only once the OAuth connection is established.

Zendesk

Salesforce

Note: Salesforce AI Actions are available for inbound calls only. Actions also write to Salesforce and consume your org's daily Salesforce API allowance. See "AI Actions: Setting up Salesforce actions for your AI Voice Agent" for full setup, billing, and troubleshooting details.

Native Aircall actions: Send SMS

Send SMS is a native Aircall action, not an integration-powered one. There's nothing to install, and any company with SMS-eligible numbers can use it.

Send SMS lets your AI Voice Agent text the caller during a call, for example a booking link, a portal or form link, a tracking link, or a confirmation once another action completes.

Adding it to an agent:

• No integration step. Add Aircall — Send SMS to your agent in the AI Actions panel directly. There's no platform-level capability to switch on first.
• Add one, refer to it multiple times. You can give an agent several Send SMS actions, each with its own name and purpose, say, one for a booking link and one for a post-action confirmation.
• No inputs to configure. Unlike actions such as Get Order Details, there's no Intake information to fill in on the action panel.
• Sender. Choose the sending number from your company's SMS-eligible numbers. If your agent is already assigned to an SMS-eligible number, it's prefilled for you. If your agent has no assigned number, or its number isn't SMS-eligible, nothing is prefilled and you choose from the list.
• Recipient. The message goes to the caller's number by default and isn't configurable at setup. It can be changed after confirmation or a failed send.
• Message content. There's no fixed template. You decide when it sends and what it says by describing it in Conversation guidelines; the agent composes the message at send time.

Sent messages appear in your Aircall Workspace like any other SMS.

Note: Send SMS fires and continues the call. It doesn't report back to the agent whether the message sent successfully, and nothing is surfaced to the caller either way in this version. Avoid writing a Conversation guidelines rule that has the agent confirm the send to the caller, since the agent has no way to know.

Step 1: Enable capabilities on an integration

AI Actions follow a two-step model. First you enable the AI Action at the integration level - this makes it eligible to be used. Then you add it to one or more AI Voice Agents - this is what actually lets an agent reach for it on a call. Both steps are required.

Steps:

1. Go to Aircall Dashboard > Integrations > Apps 
2. Open the connected integration you want to configure, for example Shopify, HubSpot, or Zendesk
3. Navigate to the AI agent capabilities section within the integration settings
4. Toggle on the capabilities you want to make available. Examples by integration:
   • Shopify: Order status updates
   • HubSpot: Enable Scheduling, Enable Contact details, Enable Company information, Ticket management
   • Zendesk: Enable Contact details, Ticket management

Important: All capabilities are off by default. They must be explicitly enabled per integration before they become available to assign to an agent.

Note: If you have multiple installations of the same integration (for example, two Shopify stores), each installation has its own AI Actions settings. Enabling an action on one installation does not affect the others.

Step 2: Assign capabilities to an AI Voice Agent

Once capabilities are enabled on an integration, you assign them to the specific agents that should use them.

Steps:

1. Go to Aircall Dashboard > AI Voice Agents 
2. Open the agent you want to configure
3. In the agent settings, scroll to Functional abilities and click AI actions.
4. Click + Add action to open the side panel.

5. Open the Action dropdown and pick the action you want to add. The dropdown groups actions by integration. Each option shows a short description of what the action does.

6. Review the data contract that appears below: Preset instructions (what the action does), Intake information (the details the agent will collect from the caller, with required fields starred), and Information available to use (the data the agent can reference once the action returns). For a small number of actions, you'll also see a Use for Caller Context checkbox here, please see Running an action before the call starts below.

7. Optionally, write rules in Additional instructions to control how the agent handles this specific action (covered in the next section).
8. Click Save action at the bottom of the side panel.

Tip: If the action you want is missing from the dropdown, and it's an integration-powered action, the toggle on the integration is probably off. Go back to Step 1 and enable it. Native actions like Send SMS don't have a toggle, so if one of those is missing, check that it hasn't already been added to the agent

During a live call, the agent decides whether and when to use each enabled AI Action based on the caller’s intent and, if specified, the rules you have written. Those rules are the subject of the next section.

Running an action before the inbound call starts

For a small number of AI actions, you can have the agent run the action automatically before the caller even finishes hearing the greeting, instead of waiting for the agent to decide mid-conversation.

This is available on the following actions:

• Search Contact in Zendesk
• Get Contact in HubSpot
• Find Customer in Salesforce

To turn this on, check Use for Caller Context on the action's panel when you add or edit the AI Actions of your AI Agent. The action runs using the caller's phone number as soon as the call starts, so the agent already has the matching contact or customer record by the time it begins speaking.

With this on, callers don't get asked for their name or email right away, since the agent may already know who they are, transfers to another agent or a human carry over better context, and calls tend to resolve faster since the back-and-forth at the start of the call is shorter.

Shaping how the agent uses AI Actions

Assigning an AI Action enables it, but defining rules improves accuracy and consistency. You can guide behavior in two places.

Option 1: Additional instructions (on a specific AI Action)

Additional instructions live inside the action panel itself. They apply only when that one action is being used, on this one agent. This is the right place for rules that are tightly scoped to a single action — how to gather inputs, which output fields the agent should focus on, what to confirm before executing.

Use Additional instructions when you want to:

• Tell the agent to collect a parameter silently, e.g. reuse the phone number from the call instead of asking for it again.
• Add a parameter that should always be applied, e.g. a default pipeline or status.
• Restrict which output fields the agent should reference in the response.
• Tighten verification for a higher-stakes action, e.g. require both email and phone before sharing detail.

Note:
• Always ask for order name and use the phone number from the call silently
• Only ask for phone number or email if no match was found

Option 2: Conversation Guidelines (across all actions on the agent)

Conversation guidelines for these actions live below the Actions in use list and apply across every AI Action on this agent. The box only appears once you have added at least one action. This is the right place for the business logic of the call - when each action should be triggered, how multiple actions chain together, and how to react to specific data the action returns.

Use Conversation guidelines when you want to:

• Define the trigger phrases that should make the agent reach for a specific action.
• Sequence multiple actions on the same call (e.g. identify the caller with Get Contact, then look up their open tickets using the userid with List Open Tickets).
• Branch on a specific return value (e.g. "If financial_status is PENDING, …").
• Describe a fallback path or transfer condition.
• Coordinate AI Actions with other call-handling rules, like Call transfers.
• Reference an action in the rules by surrounding its name with single quotes, e.g. ‘Get Order Details’ or ‘Search Contact’.

Example: Conversation guidelines on an agent with Get Order Details added:
 

Note: When the caller asks about shipping updates, package status, order details or similar:
1. Use ‘Get Order Details’
2. When you find an order read back the details of the order
3. If Payment Status is Pending, tell the customer that payments can take 3 to 5 business days (or longer) to clear depending on the provider.
4. If the Payment Status is Partially Paid, transfer the call to a human.

Conversation guidelines auto-save as you type. There is no separate save button.

Choosing the right option

Both options accept the same plain-English style. Treat each line as a rule for the agent. Be specific about the trigger, the data, and the desired response. Avoid contradictions - if two rules disagree, the agent will pick one and ignore the other.

Important: Pick one place per rule. Putting the same rule in both Additional instructions and Conversation guidelines does not make it stronger - it makes the agent harder to debug when behaviour drifts.

Testing your AI Actions

Test your setup in two stages before going live.

1. Test the action

The Testing tab on each integration page includes a Playground where you can pick an action, fill in parameters, and click Execute. No agent, no minutes consumed. Use it to confirm the connection works and the response contains the fields your rules reference.

Use the Testing tab in the integration:

• Open the MCP Playground
• Select an action
• Enter parameters
• Click Execute

This verifies connectivity and response structure.

2. Test the agent

Use Test call in agent settings:

• Validate real call behavior
• Test different scenarios such as success, alternative outcomes, and failures
• Confirm multi-action sequences

Note: Test calls consume AI Voice Agent minutes and may create real records. Use test environments or dummy data when possible.

Common issues include:

• Conflicting rules
• Poorly defined triggers
• Incorrect field references

Billing

AI Actions are included at no additional cost. Capability invocations count against the AI Voice Agent minutes already included in your plan. There is no separate billing line and no per-invocation charge.

If your AI Voice Agent minutes are exhausted, the agent will no longer handle calls and will not be able to invoke capabilities until minutes are replenished.

FAQs

Do I need to configure capabilities separately for each AI Voice Agent?

Yes. Capabilities must be enabled at the integration level first, then assigned individually to each agent that should use them. Enabling a capability on an integration does not automatically assign it to any agent.

Can I enable capabilities for some agents but not others?

Yes. Each agent has its own AI Actions section. You can assign different combinations of capabilities to different agents depending on their purpose.

Can the same AI Action behave differently on different agents?

Yes. Both Additional instructions and Conversation guidelines are stored per agent. The same AI Action can have different rules on different agents, and editing them on one agent does not affect any other agent.

Are AI Actions available on all Aircall plans?

AI Actions are included within your existing AI Voice Agent minutes. No additional SKU or add-on is required. Confirm with your account manager if you are unsure whether AI Voice Agent minutes are included in your plan.

Should I write my rules in Additional instructions or Conversation guidelines?

Additional instructions are best when the rule is scoped to a single AI Action — how it gathers input, which output fields the agent should focus on, what to confirm before invoking. Conversation guidelines are best when the rule is about when to use an action, what to say once data comes back, sequencing multiple actions, or coordinating with transfer rules. The comparison table above lays out the trade-offs.

What is the difference between enabling a capability on an integration and assigning it to an agent?

Enabling a capability on an integration makes it available in the system. Assigning it to an agent allows that specific agent to use the capability during calls. Both steps are required.

Where can I see the Conversation guidelines box?

Below the Actions in use list on the AI actions tab of the agent. The box only appears once you have added at least one AI Action to the agent.