What an AI voice agent is
An AI voice agent is a program that talks to people on the phone. A caller dials a number, the agent picks up, and the two have a spoken conversation. The agent can also place calls: it dials a contact, introduces itself and carries out a task.
What makes it an agent rather than a phone menu is that it is not following a fixed script of options. It is given a goal, instructions and tools, and it works out what to say from what the caller actually said.
How a call works
Every turn of a call runs the same loop. Speech recognition turns the caller's audio into text. A language model reads the conversation so far, with your instructions, and writes the next reply. A voice engine turns that reply back into audio and streams it down the line.
Dialgem starts speaking the first sentence while the model is still writing the second, so the pauses between turns are shorter than a system that waits for the whole reply. You choose the voice, the model and the instructions per agent.
- Listen: the caller's speech is transcribed as they talk.
- Decide: the model chooses a reply, or calls a tool such as a booking lookup.
- Speak: the reply is synthesised and played back in the agent's voice.
Why turn-taking matters most
People interrupt each other constantly. An agent that keeps talking over a caller who has already answered sounds broken, and callers hang up on it. Dialgem agents stop mid-sentence when the caller speaks and hand the floor back, which is the behaviour most callers judge the whole product by.
What a production line needs beyond the demo
A demo only ever shows the happy path. Real calls include someone asking for a person, an answering machine picking up, and a phone menu on the other end. Dialgem handles each of these.
- Transfer to a human, cold or warm, with the agent briefing the person first if you want.
- Answering-machine detection, then hang up or leave a message in the agent's own voice.
- In-band DTMF, so an agent that reaches a phone tree can press the digits it asks for.
- Call recording that never blocks the audio, served through signed links.
- A console to build the agent and call it from your browser before you connect a number.
What AI voice agents cost
A call is priced as four parts: the platform, a voice, a language model and a phone line. Across the choices on offer that comes to $0.045 to $0.135 per connected minute, billed per second from prepaid credit. There is no seat licence, no per-agent fee and no charge for a number that is not on a call. The pricing calculator itemises the bill for your volume.
Questions
- Is an AI voice agent the same as a chatbot?
- No. A chatbot answers typed messages. A voice agent handles a live phone call, which adds speech recognition, a natural-sounding voice, turn-taking and everything telephony brings with it: transfers, voicemail and phone menus.
- Can it make outbound calls as well as answer them?
- Yes. A number can have an inbound agent, an outbound agent, or both, and outbound calls can detect voicemail and leave a message.
- Do I need my own phone carrier?
- No, you can use a Dialgem line. If you already have a carrier you can bring your own SIP trunk, and calls on it carry no Dialgem line charge.