An AI agent with a phone number can call a business, confirm an appointment, return a requested callback, or answer questions on a dedicated line. AgentLine provides the US or Canada number and hosted voice conversation. Your existing agent can manage calls through MCP or REST and retrieve the transcript afterward.
These are implementation examples, not measured customer results. Start with one workflow and test it on your own phone before using it with other people.
1. Call a restaurant about a reservation
Give the phone agent the restaurant's number, date, local time, party size, name, and acceptable alternatives. For example: "Ask whether there is a table for four on Friday at 7 pm. If unavailable, ask about 6:30 or 7:30. Do not accept a deposit or another date. Read back the agreed details."
Retrieve the transcript and report whether staff confirmed a booking, offered an alternative, or could not help. A completed call alone is not a confirmed reservation. Start with the dinner reservation template.
2. Confirm or reschedule an appointment
Provide the existing appointment time, timezone, callback number, and allowed alternatives. A phone agent can ask for confirmation and record a requested change. To write a new time into a calendar, connect that calendar through your own runtime and check availability before committing.
Use the appointment reminder template for outbound confirmations or the scheduling template for a line people can call. Test timezone handling and ambiguous replies such as "next Thursday."
3. Return a requested callback
Use the callback request as the context: who asked, the reason, and the time they requested. Give the agent a short goal and a fallback if nobody answers. For example: "Call Alex about the quote they requested. Ask which of the two options they want to discuss. If they want a person, record the best callback time."
Store the result against the original request in your CRM. The GTM template is a starting prompt; your integration supplies the customer data and writes the result.
4. Answer calls after hours
Set the business hours, services, and questions the agent can answer. For requests outside that scope, collect the caller's name, reason for calling, and callback details. Only promise a callback time if your team has agreed to it.
Start from the after-hours receptionist. Verify that the transcript and follow-up task reach the person who will handle them.
5. Look up an order during a call
The phone agent can ask for the information your business uses to verify an order and request a lookup from your runtime. Connect your order system through the relay or a webhook. Keep the answer limited to the data that system returns.
The order-status template supplies a conversation outline. Test an unknown order, a delayed order, and an unavailable lookup service. Tell the caller when information cannot be verified.
6. Collect feedback by phone
For customers who expect the call, keep the survey short: one rating and one follow-up question. Provide an introduction that explains the purpose and let the person decline. Save the transcript and the response rather than assuming silence means satisfaction.
The feedback template provides a starting point. Your application decides who to call, when to call, and where to store the result.
7. Call your personal agent to capture a task
A dedicated number can let you speak a task while away from your computer. Connect the runtime that has access to your task list, then verify that it created the intended task. Register the owner number and follow the skill's owner-call setup.
Start with the personal assistant template. A phone number reaches the AgentLine phone agent; it does not automatically connect to a ChatGPT or Claude conversation. Tool access and a persistent integration determine which private tasks the runtime can perform.
Choose the runtime you already use
The same phone API can be used from a compatible MCP client, a coding agent, or a custom application. Use the specific setup guide for Claude Code, Codex, Nous Hermes, OpenClaw, Grokbot, or OpenAI Dots. Host permissions and available tools still apply. See all runtime guides.
Setup and cost
Follow the AgentLine skill to authenticate, create a phone agent, fund the account, and provision a US or Canada number. Set a system prompt that includes the goal, known facts, and what the agent may agree to. Make a test call and read its transcript before connecting live customer data.
Plans start at $20/month with one number and 100 minutes included. Calls beyond the plan are $0.12/min, billed per second after a one-minute minimum. For example, one number plus 100 connected two-minute calls is $3.00 + $24.00 = $27.00, before any texts received by the number. Your runtime or other connected services may have their own charges. See current pricing.
API reference and next steps
Use the live OpenAPI schema for request fields and the documentation for API examples. Call creation starts the conversation; the transcript provides evidence of what was agreed. For a workflow with a calendar, CRM, or order system, test the integration's result as well as the call itself.