Google Sheets
append_to_google_sheet The agent adds a row to your spreadsheet — requests, contacts, bookings.
Connects to what you already use
Four actions, a store, payments, a calendar, and email are set up with a request in the conversation. Everything else you describe yourself — in words or with a link to your service's documentation.
Set up with a request in the conversation
No code, no keys in form fields, no separate settings screen. You say what should happen — the Builder sets up the action and immediately offers to test it on an example.
SkyService CRM
create_skyservice_client Adds the customer to your database: name and phone from the conversation go straight into the CRM.
Telegram
notify_manager_telegram Notifies a manager about a new request or an important question right in the chat.
Zapier, Make, n8n
send_webhook A webhook to your workflow — and from there to HubSpot, Notion, Slack or your own backend.
Store, payments, calendar and email
Also with a request in the conversation: the Builder will tell you where to get the key or link and check that it can be read.
WooCommerce, Horoshop, Prom.ua
“Where's my order?” The assistant reads the order from your store itself and tells the buyer the status and tracking number. Only to someone who gives the order number and their phone or email.
Store product file
The assistant knows your whole catalog: suggests alternatives, gives current prices and stock. The platform finds the file on your website by itself.
Payments via LiqPay
“I'll take it — how do I pay?” The assistant sends the buyer a payment link right in the conversation and knows whether it's been paid. The money goes to your store's LiqPay account: you're the seller, and the platform never touches the payment. Refunds are handled in your LiqPay dashboard; the assistant doesn't issue them.
Calendar
Google, iCloud, Outlook — with a secret link it sees when you're busy, and with an app password (iCloud, Fastmail, Nextcloud) it can also add events.
A support inbox the assistant uses to answer customers, or your personal one, which it only reads when you ask.
And if your service isn't listed
Then the same approach works, just with more work — and you do it, not us.
Your API
If the service has documentation, the agent works with it on its own — following the description you gave in words or as a link.
This part takes some work
You describe the action you need in plain words right in the conversation — or share a link to your service's docs, and the assistant will pull the list from there itself. If someone else set up the system for you, this is the moment to bring them in. We don't pretend it connects by itself.
Once the assistant creates an action, it immediately offers to run it once with sample data. Say yes: an untested action is a bug that will only surface in a conversation with a customer.
What the agent sees from your service's response
As much as you allowed it — and no more.
Keys never end up in the conversation
Tokens and passwords are stored separately from the action text, filled in at call time, and never appear in the model's replies.
Responses are read as data, not as commands
Whatever someone else's service returns, to the assistant it's a response to a request, not a new instruction about what to do next.
Every call is visible in the dashboard
The action's name, time and effect appear in the conversation next to the message that triggered it — and in the change log, where it can be undone.
And if you came with code
The same thing, from the other side: a full API behind a key, an MCP server for your AI assistant, and outgoing webhooks.
Developer page — what a key can do, how to connect Claude or ChatGPT to the Builder, and which events the platform sends on its own.
Tell it where to put requests
In one sentence: “Log requests from the website in my Google Sheet”. Then the assistant sets up the action and tests it on an example.