Early beta · launching soon
Callmark puts an inaudible, signed ID inside AI voice audio, so a bank or a caller can check who sent the agent.
Verified agent
Acme Bank collections
A cloned voice costs almost nothing to make now. Caller ID can be spoofed. Any metadata attached to an audio file gets stripped once the call crosses a phone network. So when an AI agent calls a customer, the person on the other end has no way to tell whether it is the real one.
Companies that run voice agents need to be able to prove an agent is theirs. The people and systems receiving those calls need to be able to check that while the call is still going.
Callmark embeds a compact, machine-readable ID into the audio as the agent speaks. Because it isn't a header or a tag, a codec or a carrier has nothing to strip.
The record says which company registered the agent, what the agent is authorized to do, and when the audio was generated. A receiver can use the signature to confirm nobody has altered the record.
The same ID tells a listener that AI generated the voice. Rules such as Article 50 of the EU AI Act are pushing companies to label synthetic audio, and Callmark gives you a machine-readable way to do that. On its own it does not certify compliance.
You can use the hosted API, or run the SDK in your own cloud so call audio never leaves your environment.
These are goals for the system. We have not measured them yet.
You create an identity for each voice agent, covering your company, the agent's purpose, and what it is allowed to do. Callmark signs it and gives it an ID.
The SDK sits in your text-to-speech pipeline and writes the ID into the audio as it is produced. Your agent sounds the same as before.
A bank, a contact center or a carrier checks a few seconds of audio, either through the API, through the SDK, or inside their own network.
The ID resolves to the signed trust record, which holds evidence a fraud team or an auditor can keep.
Values shown are sample data.
Criminals copy collections calls, fraud alerts and account callbacks more than any others. Mark your outbound agents so that customers and your own fraud team can tell a real call from a cloned one.
Inbound lines now get calls from AI agents as well as from people. Check which company sent an agent before it gets anywhere near an account.
Appointment reminders, claims follow-ups and benefits calls carry personal data. Give patients and members a way to confirm a call is legitimate.
Direct access to the founder, help wiring Callmark into your pipeline, and a say in what we build next.
Honest feedback, a short call every couple of weeks, and real audio paths to test against.
Access starts with a conversation. There is no signup form. Write to the founder.
We haven't set pricing yet. Early partners agree terms with us directly.
It is designed to be inaudible. Part of the beta work is testing that across voices and phone lines.
If you run the SDK in your own cloud, audio stays in your environment. With the hosted API, we process the few seconds needed to check an ID, and we agree retention with each partner in writing.
One pipeline in the beta today. Tell us what you run and we will give you an honest answer on how far away it is.
No single tool does that. Callmark gives you a machine-readable way to mark synthetic audio, which covers one part of what Article 50 asks for.
When the ID holds up across the phone networks our early partners use. We will share a date once we have one we can keep.
Voice is becoming the way AI agents reach people, and it is the easiest channel to fake. We built Callmark so that a person on the phone, and the systems around them, can trust the voice they hear.
We think proof of who sent an agent belongs inside the audio, where it travels with the call, and not in metadata that gets lost along the way. We are building that layer with a small group of early partners.