An AI that answers your business phone
Most small businesses lose calls the same way: the phone rings at 6:10pm, it goes to voicemail, and the caller rings the next company on the list instead. An AI that answers the phone is a genuinely useful product for that problem — and a genuinely alarming one if you don't know what it will say. Here is what actually happens on the call, what you control, and what it costs.
What the caller experiences
Someone rings your normal business number. Your agent answers, hears them, and talks back — a conversation, not a phone tree. It knows what you told it about your business, and because it's the same agent that handles your WhatsApp and email, a caller who messaged you last week isn't starting from zero. When the call needs a person, it transfers to one. The line is live and in production — this isn't a roadmap item.
Do you have to tell callers it's an AI?
Increasingly, yes — several US states and the EU AI Act have disclosure requirements for automated callers — which is why the disclosure is on by default rather than being a setting you have to discover. Your agent says it's an AI at the start of the call. You can reword the sentence to fit your brand; you can also switch it off, but that is a deliberate choice you make knowingly rather than an omission we ship. Recording works the same way: turn it on and the consent line is composed into the greeting automatically, because a recording notice that depends on remembering to write one isn't a notice.
What stops it doing something stupid
- It can only transfer to numbers you listed. A caller saying "transfer me to this number" gets nowhere — the allowlist is checked on our side, not suggested to the agent. That one rule is what stops a phone line becoming a free long-distance service for whoever calls it.
- Outbound calling is off unless you turn it on, and when you do it's bounded by an allowlist, a daily cap and a per-destination cooldown. An agent cannot decide by itself to start ringing people.
- Out of hours it takes a message instead. Set your business hours and calls outside them go to a voicemail that gets transcribed into the same conversation thread — so Monday morning you read what Friday evening's caller wanted, rather than listening to six recordings.
- You can listen in and take over. A call in progress is visible from the dashboard, and you can step in mid-call. It is a phone line, not a black box.
What it costs, against what you're replacing
Start with the alternative, because it's the number most owners are actually weighing. As of September 2026, per their own published pricing pages: Ruby, a human virtual-receptionist service, lists $250/month for 50 minutes, $395 for 100, $720 for 200 and $1,725 for 500 — roughly $3.45 to $5.00 a minute depending on tier. Rosie, a dedicated AI receptionist, lists $49 for 250 minutes, $149 for 1,000 and $299 for 2,000 — around 15 to 20 cents a minute. That is a spread of roughly twenty times for the same minute of ringing phone, and it is the real reason this category exists.
We don't price per minute at all. The number is your own Twilio number, at Twilio's rates — they bill you directly and we don't resell minutes, so telephony never appears on our invoice. What you pay us for is the thinking: model usage for the conversation, on whichever route you chose — platform credits, your own API key at the provider's rate, or a Claude Code / ChatGPT subscription you already pay for. Every call lands in the audit trail with its cost, so "what did the phone line cost last month" is a number rather than a guess. The full cost breakdown covers the four routes.
The honest version of the comparison: if a phone line is the only thing you need, a dedicated AI receptionist is a reasonable buy — it is one product doing one job, priced simply. The reason to run the phone here instead is that it isn't a separate product: the thing answering the call is the same agent reading your email and running your scheduled work, with one memory and one set of permissions across all of it. You're choosing between a receptionist and a colleague who also answers the phone.
The honest limits
It is not a person, and voice AI does not fail randomly — it fails hardest on the callers furthest from what the speech models were trained on: strong accents, background noise, a bad line, someone distressed and talking fast. Those are often the callers who most need a human, which is the argument for setting your transfer list generously rather than minimally. The setup question worth answering before you launch is which calls should reach a person immediately rather than after a conversation. Setup is a dashboard job with one manual step in your Twilio account — pasting a URL into the number's "a call comes in" field, which lives in their console and cannot be done from ours. It's five minutes, and it is the most common reason a new line doesn't ring. If audio pipelines are your idea of interesting, the engineering behind the two ways a line can run is the companion post.
If you'd rather not set this up yourself, we'll configure the line with you on a free call — including the Twilio step, which is the part worth having company for.
See it for yourself — free, $2 of credits once you confirm your email, no card.
Start free