What it connects
Integrations are what let the agent act: look up an order, issue a refund, create a ticket, notify your team. You connect your tools once and the agent can use them through the actions you turn on.
- Commerce and paymentsShopify for orders, and Stripe for refunds, subscriptions and invoices.
- CRM and ticketsHubSpot, Zendesk and Intercom.
- Your teamSlack, Jira or Linear to send alerts or open issues.
- Calendar and docsGoogle Calendar for appointments and Notion for knowledge.
- ChannelsWhatsApp connects right from the console.
Your own API
If your system isn’t on the list but has an API, it works too. You describe the action — what it does, which URL it calls, what data it needs and how it authenticates — and the agent uses it like any other.
- 1You describe it. A title and a clear description, which is what the agent reads to know when to use it.
- 2You set its risk and permission. Whether it only reads or also writes, and whether it runs on its own or with approval.
- 3You test it. A real call to your API before letting it serve customers. If it writes, you have to confirm.
How it’s protected
- Sealed credentialsKeys are stored encrypted and never shown again.
- Public addresses onlyHTTPS and no internal addresses: the agent can’t call your private network.
- No double chargesThe same action with the same data in the same conversation doesn’t run twice.
- Smart retriesReads are retried; writes only if the request never went out.
A write only counts as done once it has been verified by reading the system back. Without a connected integration, the action is a simulation and is never treated as done.
