Some errands still only happen by phone. A plumber who never answers email, a clinic that confirms bookings by voice, a supplier who quotes only when you ring. Each call can take ten minutes of dialling, waiting on hold and writing down what was said. CALL-E is a tool that hands those calls to the AI assistant you already use, and it launched on Product Hunt this week.

What it is

CALL-E is a phone-call agent that your AI assistant can use to place real outbound calls. You describe the errand in chat, it shows you a call plan, and it dials only after you confirm. According to the company, it handles phone menus, hold music, voicemail and transfers, and every result comes back with a transcript, the evidence behind it and a confidence label.

It plugs into the assistants many builders already run. The homepage names Codex, Claude Code, Cursor, OpenClaw and Hermes, plus a ChatGPT App Store listing and WhatsApp. For developers there is an MCP server (MCP, the Model Context Protocol, is the standard way to give an AI assistant a new tool), TypeScript and Python SDKs, and a REST API. The changelog shows a fast pace: MCP V2 and a ChatGPT plugin on 19 October 2026, HubSpot, n8n and Dify integration updates on 26 October, and SIP integration (SIP lets you route calls through phone lines you already own) on 2 November.

Who it's for

It suits solo operators and small teams in the US or Canada who lose hours each week to short, repetitive calls: confirming appointments, rescheduling, checking stock, asking for quotes. It also suits builders who want to add calling to an agent or an automation without wiring up telephony themselves. If your customers and suppliers are elsewhere, read the Limitations section before you sign up.

What it actually does

A realistic case: you run a small physio practice in Austin and have 20 patients booked for tomorrow. Missed appointments cost you money, so you want each patient called, the time confirmed, and anyone who cannot make it offered a new slot.

In the chat version, you tell your assistant something like: "Call tomorrow's 20 patients, confirm their time, and if they cannot come, offer Thursday at 10am or 2pm." Fill in any gaps, such as how to introduce the practice, before you confirm; we covered why a clear brief matters in our piece on agents that ask questions. You check the plan, confirm, and the calls go out.

For a job you will repeat every day, the better route is a saved template the company calls a Goal. You write the call once, publish it, and then run it against each phone number with a few variables (patient name, appointment time). The billing docs recommend Goal mode for batch calling and describe one-off calls as meant for development and testing.

The most useful part is what comes back. Through the API you send a result schema, which is a short description of the fields you want filled in. For our physio example that might be three fields: confirmed (yes or no), new_time (text) and notes (text). The quickstart says the result is checked against that schema, and a status field tells you whether a valid result is ready, still pending, or unavailable. That means you can drop the answers straight into a spreadsheet or CRM instead of reading 20 transcripts.

The same pattern works for supplier quotes: one Goal that asks for price, lead time and minimum order, run across five suppliers, with the answers returned as rows you can compare.

Pricing

Pricing as of 15 Nov 2026 is pay as you go in US dollars, and new users get 100 free credits. Each call costs a model fee plus a carrier fee. Per the billing page, a one-off US or Canada outbound call costs $0.0296 per 10 seconds of AI time plus $0.04 per minute for the phone line. In Goal mode those halve to $0.0148 per 10 seconds plus $0.02 per minute. The model fee is billed in 10-second blocks and the carrier fee is rounded up to whole minutes.

The company's own examples: a 30-second call costs $0.1288 one-off or $0.0644 with a Goal. A 60-second call costs $0.2176 one-off or $0.1088 with a Goal.

Worked example (illustrative): our 20 physio confirmations, if each runs about a minute in Goal mode, come to 20 x $0.1088, roughly $2.18. The same calls placed one-off would cost about $4.35. Set that against the time saved: at ten minutes a call by hand, 20 calls is over three hours of reception time. Two caveats from the billing page. Model fees may apply even if a call does not connect, so a list full of wrong numbers still costs something. And an experimental success fee is listed but currently fully waived, so watch for changes.

Getting started

  1. Sign up on the CALL-E site and claim the free credits.

  2. Connect it to your assistant: the ChatGPT app, the official calle skill for coding agents such as Claude Code or Codex, or the MCP server.

  3. Place one test call to your own mobile. Listen to how it introduces itself and check the transcript.

  4. Write your repeat errand as a Goal, with variables for the parts that change on every call.

  5. If you use the API, create a key in the dashboard and keep it on your server, as the docs instruct. Our note on scoping credentials for agents applies here.

  6. Run a batch of five, read every transcript, then scale up.

Alternatives

Three alternatives are platforms for building your own voice agent, which means more setup and more control.

  • Retell AI: lists $0.07 to $0.31 per minute for voice agents, with $10 in free credits. Pick it if you also need inbound calls answered and want templates and testing tools.

  • Bland AI: its Start plan has no monthly fee and charges $0.14 per connected minute. Pick it for high-volume calling where you will tune the agent yourself.

  • Vapi: charges $0.05 per minute for hosting, with speech, model and phone costs added on top. Pick it if you have a developer and want to choose every component.

Decision rule: if the job is errands you would otherwise do yourself, started from a chat, CALL-E is the shorter path. If you are building a phone product for customers, start with one of the three above.

Limitations

  • North America first. Local numbers cover only the US and Canada. The regions page lists 20 more countries, including Australia, the UK and India, on international numbers meant primarily for development and testing.

  • Outbound only. The homepage says it does not pick up or screen incoming calls. The billing page lists inbound calls for Goal mode only.

  • Line capacity. The shared line has a cap on calls running at once, and the error docs point shared-line users to identity checks (KYC) and buying a number to raise it. Big batches need that step.

  • Hard to predict cost. Billing in 10-second blocks plus rounded minutes makes a budget awkward to estimate, and the team agreed on Product Hunt that a price calculator would help.

  • Old API retiring. Legacy call tasks are planned for retirement at the end of 2026, so anyone on the old SDK needs to migrate.

An AI voice is treated as an artificial voice in the US. In February 2024 the FCC adopted a declaratory ruling confirming that the Telephone Consumer Protection Act's limits on "artificial or prerecorded voice" calls cover AI-generated voices, so such calls need the called party's prior express consent unless an exemption applies. CALL-E's terms of service put that duty on you: you must obtain legally required consents, clearly disclose that a call may involve an AI voice, and avoid numbers on Do-Not-Call lists where contact is prohibited. A practical checklist: call only people who gave you their number for this purpose, have the agent say it is an AI assistant calling for your business in its first sentence, and get advice before calling consumer mobiles in bulk.

Verdict

Try it if phone errands or appointment confirmations eat your week in the US or Canada; skip it if your contacts are outside North America.

Comments

Avatar

or to participate