Voice identity, answered.

Vosco is voice identity for AI that acts. It gives AI systems a real-time trust signal for voice — who is speaking, whether the voice is live, generated or replayed, and whether an action should be allowed — via a cloud API for voice agents and an on-device SDK for robots and intelligent devices.

In a single call: synthetic / cloned-voice detection, replay and liveness, speaker identity (1:1 verification and 1:N identification) and a combined trust score with a policy decision — plus insights including gender, age range, sentiment, language and audio quality.

The signals combine into one trust score. You define what each action requires, and Vosco returns an outcome your AI can act on: allow, challenge, deny or escalate.

Both — and both are available now. They are two deployment environments for the same trust technology: the Vosco API (cloud or on-prem) for voice agents and automated workflows, and the Vosco SDK (on-device) for robots, smart devices and vehicles.

Yes. Vosco Nano is a 7M-parameter model that runs fully on-device on iOS, Android and embedded hardware — under 10 ms on an NPU, 100% offline. No audio has to leave the device.

For higher-risk actions, a trusted personal device — a phone, watch or earbuds — can independently observe the same interaction, verify the enrolled speaker and give the machine a signed trust attestation, locally and without an internet round trip. It is an independent verification signal for higher-assurance actions, added only when an action needs it.

We build the layers that decide who is speaking and whether to act — speaker recognition, trust and insights — and partner for commodity layers such as ASR and TTS. Robotics teams integrate Vosco as the voice identity layer in their existing stack.

Vosco listens continuously and returns a fresh verdict every 3 sec, with inference latency under 10 ms on GPU or NPU — so it keeps confirming it is the same live caller from the first words to the last, before every action rather than after it.

Vosco Precise reaches 99.3% speaker accuracy; the on-device Vosco Nano reaches 98.4%. Both lead the efficiency index — synthesis-detection accuracy per million parameters — against leading commercial vendors. See Research for the full table.

New voice models ship every week. Vosco continuously retrains against the latest open-source and commercial speech models, with an automated research pipeline in progress.

Send audio over REST or WebSocket, call the MCP server from an AI agent, or embed the Vosco SDK on the device. The response shape is the same in every environment. Start at Developers.

Yes. The Vosco MCP server exposes verification as a tool to Claude, Cursor and any MCP-compatible agent. Platform integrations for Vapi, Twilio, LiveKit, Pipecat and Recall.ai are being built with design partners — see Voice agents.

The API is usage-based at $0.0042 per minute, lower at volume, with 1,000 minutes free every month. Enterprise on-prem is an annual licence; the Vosco SDK is licensed per device, per year. See Pricing.