Meet Oxylabs at VapiCon 2026 in San Francisco

Voice agents are easy to impress with, yet hard to keep reliable. A demo may go smoothly if the speaker is clear, the room is quiet, and the script is predictable. But callers often mumble, switch languages mid-sentence, or even talk over the agent.

Most of the gap between those two worlds comes down to data: how much of it a model has heard, how varied it is, and whether you have the right to use it.

That's the problem we're taking to VapiCon on November 11-12, 2026.

Why VapiCon, and why now

Voice agents for business use are no longer the hypothetical scenario they once were. Teams already have voice agents live with real customers. In response, VapiCon 2026 is inviting experts from different corners of the industry to set a sturdier base for improving them over time.

And we have an answer to what that base needs: new languages, new accents, new ways people actually talk. Helping with exactly that is what Oxylabs does.

After a decade of building our fast-adapting infrastructure, we can offer voice AI teams three routes: ready-to-use Video Datasets, creator-approved with verified consent; the new Web API for dependable access to the platforms you need; or high-bandwidth proxy solutions to collect your own data at scale.

Who you'll meet

A good mix of our data experts is coming to see how this sector could benefit from public web data collection solutions. So if you're trying to work out where your next dataset comes from, you can talk it through with someone who helps deal with such problems on a daily.

At the stand, you can find solutions to the problems that come up in nearly every voice AI project:

• Training for more languages and accents. Our Video Datasets include 4M original videos from 1M channels, all creator-approved with verified consent for AI training. Video comes in 720p, with higher quality available on request.

• Making speech sound natural. The Web API downloads video and audio from platforms like TikTok and Bilibili, along with subtitles and metadata, so you can work with real conversational speech.

• Grounding agents at scale. Collecting data at scale takes serious infrastructure. High-Bandwidth Proxies handle 200Gbps+ and work with yt-dlp, while Datacenter Proxies offer 99.9% average uptime and a 0.222s average response time.

Find us at the stand

We're at Festival Pavilion for the whole conference. Come by our booth any time and bring a real use case, even a half-formed one. We'd rather talk about your actual setup than give a pitch.

Make sure to come to the demos, where we show the tools running. You're welcome to challenge us with questions or just stop by to try your luck in the raffle that follows right after the practical sessions.

Always up for a discussion

If the opening of this article sounded familiar, you'll want to join this session as well. Vidas Bacevičius, our Lead Solutions Engineer, joins a panel on why agents that shine in demos struggle in production.

Production vs. Demo: How to Keep Agents Reliable

When: November 11, 3 PM | Where: Builder Track
With: Michael Grinich (WorkOS), Sourish Biswas (HOAi), Utsav Singhal (Interact AI)

Panelists will be covering what dependable agents have in common. Is it clean data, low latency, fallback logic, human handoff, or something else? Expect a highly practical conversation on getting voice AI to work in any conditions.

Discount for VapiCon

Use code OXYLABS20 for 20% off your tickets.

If you're going, find us on the floor. But if you're building voice AI and can't make it, you can always reach us at sales@oxylabs.io.