Argus understands the state of body and mind from just 15 seconds of speech and acts on it in the moment, delivering safety and experience value to customers and users. Live in deployments since January this year, it reads 30+ voice biomarkers and puts them in the context of the conversation to recommend the right action.

Voice is fast becoming the predominant way we communicate, not only with each other but also with machines. ChatGPT, Gemini and Claude all hold live spoken conversations, and this year Google folded voice into Search itself, worldwide. Meta built its smart glasses around voice. Amazon has just launched devices that listen through your day, from your wrist or the back of your phone. BMW’s 2026 cars debut an assistant running on Alexa technology, and voice-native cabins are becoming the norm across carmakers.

Behind each of these experiences sits a whole stack of technologies, built up over years: speech recognition to hear (e.g. Speechmatics, Deepgram), language models to decide, speech synthesis to reply (e.g. ElevenLabs, Rime), agent platforms to orchestrate the exchange (e.g. Vapi, Pipecat), and the infrastructure to carry it all in real time (e.g. Agora). The newest architectures already run supporting agents in the background of live calls, and several claim to read how the speaker is doing, though few of those readings rest on clinical-grade validation and rigorous science. The industry is arriving where thymia has spent six years building. In billions of conversations, the signal in the voice still goes unread: the patient playing down their symptoms, the caller who has given up by the third transfer, the driver more tired than they will admit.

A comprehension layer for every conversation, as it happens

For the first time, the complete signal in a voice conversation can be deciphered and acted on as it happens. Argus interprets the biomarkers as the situation unfolds and returns insights that become action immediately: escalating a case to a human, adjusting the cabin to the driver, redirecting an AI agent’s next response to turn the call around, delivering safety and experience value that was previously lost.

Inside Argus sits an elaborate engine for turning conversation into action: audio streams in, Helios extracts the biomarkers in real time, a policy tuned to the use case interprets them, and the recommendation goes to the partner’s system.

Argus is the foundation partners build with, wherever they sit on the stack and whatever voice context comes next. It is a horizontal intelligence infrastructure layer that augments every other layer of the voice AI stack without new hardware, so any conversation that passes through a microphone is in play. Argus draws its biomarker outputs from Helios, our real-time voice biomarker foundation model. Every biomarker is clinical-grade:

Helios’ mental health detection builds on the robust technology behind Apollo, our Class I-registered medical device product: a support tool for qualified healthcare professionals assessing depression and anxiety in adults, registered with the Medicines and Healthcare products Regulatory Agency (MHRA), and the first medical device to measure the symptoms of depression and anxiety from voice alone.

Deployed in cars and voice AI systems across the world

thymia’s technology is in production with leading automotive manufacturers and with voice AI providers worldwide. It detects the same 30+ biomarkers in every deployment and bases its recommendations on the signals that matter in that context, and how:

• Automotive: detect cognitive load and driver fatigue and recommend the in-cabin response, from a break prompt to adjusted settings, early enough to prevent an accident.

• Voice agents: give an agent awareness of how the human in the conversation actually is, so it adapts pace and approach, shapes its responses to settle an agitated caller, and escalates to a person when the signal calls for it.

• Contact centres: read caller and agent signals in every call and hand agents actionable insights mid-call, maximising customer satisfaction while the caller is still on the line.

• Healthcare: triage and augment clinical notes as people speak, from the patient calling their clinic or hospital to the consultation an ambient scribe is documenting ready for an Electronic Health Records (EHR) system, surfacing signals related to the patient’s health as a whole (not just what they are speaking about) for the care team to act on holistically.

The same biomarker plus interpretation layer applies anywhere voice is part of the experience.

Built for the real world, proven in the toughest industries

Argus runs at low latency in the environments conversations actually happen: a noisy car cabin, a busy contact centre, a phone line. thymia started where the stakes are highest, in automotive and healthcare, and holds every biomarker to the same clinical-grade standard: validation against gold-standard measures, testing for biases, and performance checked across population groups and potential confounds. Published guidance treats an area under the curve (AUC) of 0.8 and above as clinically useful, and our technology clears that bar from the very first interaction, validated with an AUC of 0.8 and above in peer-reviewed studies, across multiple languages and every major English accent. Every voice is unique, and our technology accounts for that by personalising its baseline for the individual with every interaction. Over time, you become your own baseline.

Deployment is flexible, depending on the partner’s architecture: on-cloud, on-premises or on-device via Software Development Kit (SDK). The standards behind it are certified and public: ISO 13485:2016 for our quality management system, ISO 27001:2022 and Cyber Essentials for information security, and personal data handled in line with GDPR in Europe, HIPAA in the United States and PHIPA in Canada.

“Until now, voice biomarkers meant analysing a conversation after it ended, when the moment to act had already passed. Argus is the first platform to read the biomarkers live and recommend action while the conversation is still happening, and it has been doing so in deployments since the start of the year. As voice AI takes on more of the world’s conversations, that understanding becomes essential. We think every voice system will work this way,” said Stefano Goria, Chief Technology Officer and Co-founder of thymia.

Clinical-grade, private and secure by design, built on six years of peer-reviewed science

Argus is ethical by design, consent-based and human-centric: the principles thymia builds to in healthcare, applied wherever its technology runs. Voice sits at the intersection of five major biological systems: respiratory, neurological, cardiovascular, metabolic and musculoskeletal. Change any one of them and the voice changes with it. thymia’s models pick up those changes as what we call neural morphemes®, the building blocks of biological information in voice, and combine them into the biomarkers Argus reads.

The methodology is peer-reviewed and published in Nature Scientific Reports, on the largest dataset of its kind, more than 100,000 unique voices measured over time, amongst other types of data. A deep pipeline, the result of cutting-edge research and robust engineering, keeps producing new states. These states are added to Argus and deployed only once validated, and run only with end user consent, through our own careful study design and by third parties, to uphold the highest standard.