Smallest.ai raises $13 million to build a voice AI that you cannot tell is a machine
The startup wants AI phone agents to sound so natural that callers never notice they are not speaking to a person. Here is how it plans to do that, and what it means for anyone who ends up on hold.
Key points
- Smallest.ai, founded in late 2024, raised $13 million in a Series A round led by Seligman Ventures, bringing total funding to over $21 million.
- The company builds a small, specialised voice model that listens and thinks at the same time it speaks, cutting the awkward pause callers notice with today's AI phone agents.
- If the AI cannot answer a question, it hands off to a larger, more powerful model in the background and tells the caller it is "researching" the issue.
- Paying customers already include RingCentral and Truecaller.
- Founder Sudarshan Kamath says the company's sole goal is to pass the Turing test, getting callers to finish the call without ever knowing they spoke to software.
Anyone who has called a customer support line and immediately heard that flat, slightly-too-slow robotic voice knows the problem. AI phone agents are getting smarter, but they still sound like machines. Smallest.ai, a startup launched in late 2024, is trying to close that gap.
The company just raised $13 million in a Series A round, first reported by TechCrunch. Seligman Ventures led the round, with Sierra Ventures and 3one4 Capital participating. Total funding now stands at just over $21 million.
So what makes this different from every other voice AI?
Most voice AI today runs on a large language model, the same kind of technology that powers chatbots like ChatGPT. Those models work in chunks: they wait for you to finish speaking, then think, then reply. That gap feels fine in a text chat. On a phone call, even half a second of silence sounds wrong.
Smallest.ai is building a smaller, purpose-built voice model that works differently. It listens and prepares a reply simultaneously, the way a person does mid-conversation. "While I'm speaking to you, you're already thinking, and you might interrupt me if I talk for too long," CEO Sudarshan Kamath told TechCrunch. His model is designed to work exactly like that.
When a question is too complicated for the small model, the system quietly passes it to a larger AI in the background. The caller hears something like "let me look into that for you," which is exactly what a human agent would say. No one gets cut off or left waiting in silence.
We covered a similar ambition on 15 July 2026, when Rime raised $24 million to make AI phone calls feel less like talking to a robot. Smallest.ai's angle is distinct: rather than training on better data, it bets the architecture itself, a small real-time model backed by a heavier offline one, is the fix.
What does this mean for ordinary callers?
For anyone on the receiving end, the experience is supposed to feel like talking to a patient, knowledgeable support rep. Whether that is reassuring or unsettling depends entirely on the person.
The company says it handles diverse accents, supports dozens of languages, and operates in noisy environments better than general-purpose models because voice is its only job.
| Feature | What it means for callers |
|---|---|
| Zero-lag replies | No awkward silence after you speak |
| Simultaneous listening and thinking | Can handle natural interruptions |
| Hand-off to larger model | Complex questions still get answered |
| Accent and language support | Works across many regions and languages |
| Noisy environment handling | Performs even on bad phone lines |
Current customers include RingCentral, a large business phone platform, and Truecaller, best known as a caller-ID app. Kamath says any company running a customer support operation is a potential buyer, including newer AI-first support startups.
Competitors include ElevenLabs and Cartesia, along with regional players such as Sarvam. Smallest.ai's bet is that focusing entirely on real-time phone conversations, rather than audio dubbing or podcast tools, gives it an advantage.
One honest takeaway: If you work in customer support, this technology is coming for the most repetitive parts of that job first. The practical move is to get familiar with how these AI agents are deployed so you can position yourself as someone who trains, monitors, or escalates for them, rather than someone they replace.
