Jordan Dearsley

GPT-Live is now in beta for every builder in Vapi .

OpenAI's full-duplex speech-to-speech model, live on Vapi. One model listens and talks at once. A reasoner runs your tools and skills in the background. Faster turns, cleaner interruptions, better accuracy on names and numbers.

vapi.ai/blog/gpt-live-…

Calling it now so I can brag later

1. Voice will be the default interface
2. Aliens are real
3. BCI will be as big as AI in 1-2 decades

You can feel whether the voice on a call is a human or a machine before you can explain why.

Today, @Vapi_AI is launching the Humanness Index™, a crowdsourced leaderboard for model humanness.

You are the benchmark. Cast your first vote today: humannessindex.vapi.ai

By the time you die, you will have spent 43 days of your life on hold.
Three trillion minutes a year of business-to-customer calls.
We started @Vapi_AI to end it.

LinkedIn has quietly became our #1 channel for trust, virality, and reach. So we figured it’s time to share how we got here.

This Thursday, we're hosting a live workshop: "Growing YC Founder Brand on LinkedIn"

Thursday, Vapi office, 6PM! Exclusively for YC founders. Register here: lu.ma/uj9ca9r9

Shipping today: Outbound Campaigns in Vapi.

Here’s how it works:
- Upload a CSV with phone numbers + custom variables
- Pick the assistant or workflow to run
- Choose a verified number for reliable pickups
- Launch the campaign from the dashboard

Real-time voice deserves real-time transcription. Check it out on Vapi

Join us for future events & hackathons

Shipping today: Vapi Workflows

A visual + programmable layer to design how your agents speak, think, and act - across steps

The phone is still the universal API in healthcare. Not FHIR. Not a clean REST endpoint. A literal phone call.

Most clinics, insurance carriers, and providers rely on voice as the only way to exchange data. So when people ask, “Why voice AI in healthcare?”

The answer is simple: It’s the only interface that always works. But there’s a catch, shipping here isn’t easy.

We’re not building voice agents for pizza orders. This is PHI, test results, prescriptions, triage.

At Vapi, we’ve had to build a whole new stack to make this safe:
→ Model determinism: No improvising. guardrailed, repeatable flows.
→ HIPAA mode: Zero data retention. We store nothing. Model vendors don’t either.
→ Deployment guardrails: Customers handle the PHI. We just provide the rails.

Early adopters have been digital health companies - startups building tools for systems and patients. They move fast. They care about margins. They’re willing to experiment.

Larger health orgs will follow, but trust takes time. We’re okay with that. We’re playing the long game.

Voice is the default here, not just a nice-to-have. We’re simply making it work, every time.

Vapi 2.0 is live!

Our team has been hard at work for the past few months, building on what it takes to actually get voice agents into prod.

New in 2.0:
- 100+ model integrations (transcription, LLM, TTS)
- Conversation workflows and knowledge bases
- Vapi voices that sound human
- Rock-solid enterprise infrastructure to scale to millions of calls
- Real-time observability with Langfuse integration
- A metric ton of voice agent configuration options

📍 Product Hunt: vapi.sh/4G6m0MG

📞 Try it live: 1-844-HEY-VAPI

Scheduling 30+ intake calls across time zones is a nightmare. @curious_vii solved it with voice agents.

@nikhilro_ met Christian at an event a few weeks ago, and he shared how Vapi has been a core part of his success as a solopreneur.

He runs high-intensity AI workshops and needs deep context from every participant without the calendar chaos.

His workflow with Vapi:
→ Spins up a phone line with a voice agent
→ Participants call in when it works for them
→ Agent asks custom prompts, collects transcripts + summaries
→ Data routes to Slack + HubSpot, ready to analyze

20 participants called in ahead of his latest workshop. No Calendly links. No coordination. Just async conversations and rich insights.
The result? The highest-rated session of the entire event.

This is why voice matters. It captures nuance like a human, scales like software, and removes friction from workflows that need context.

Full breakdown + video here: vapi.sh/p3yATu4

We had no idea how to close an enterprise contract… but somehow we did.

One day, a big company started poking around our self-serve dashboard. No email. No intro. Just… lurking.

At the time, Vapi was barely ready for enterprise. We didn’t even have the basics figured out.

Still, I reached out: “Need help?” That single message kicked off the most chaotic month of my life.

We scrambled to answer questions we hadn’t thought about:
- SOC-2, HIPAA, data residency?
- An SLA? We have to stay up, like always?
- How do you even write a contract like this?

I literally made up a number: “What if you used a million minutes a year?” They said YES.

We were shocked. Back then, a million minutes felt insane (now we do that every day)

The pricing was messy. The MSA was 50 pages. But that customer pushed us to grow up fast.

We learned what “enterprise-ready” actually means. They raised the bar and we had to meet it.

Since then, we’ve gotten a lot better at this. But that first one was our real training ground.

As with anything at an early startup, when there’s an opportunity, you just have to figure it out.

Asked GPT-4V to Studio Ghibli our “phone tree” photoshoot. Nailed the vibe.

But still some hallucinations… check out the mini Golden Gate and the cords. So close but not quite there yet.

Everyone’s talking about Sesame. We have it.

Traditional voice agents turn speech into text, put it through an LLM, then convert text back to speech. At each step, all the nuance is lost.

Sesame changes this. It’s a speech-to-speech model, which means it can natively hear and produce speech, just like humans do.

Voice agents built with Sesame can hear someone’s hesitance and produce a comforting response. They just feel… human.

With Vapi’s Sesame integration, you can:

- Configure agents with all the usual Vapi configuration and tooling
- Deploy Sesame voice agents in phone calls and websites
- Test, iterate, and redeploy, all in Vapi.

We’ve been waiting a long time for speech-to-speech models to hit human performance, today it’s here.

The current release from Sesame, CSM1-B, is only a voice model. We've been waiting a long time for speech-to-speech models to reach human performance, and these are still early steps.

Try integrating Sesame: vapi.sh/sesame

Last week, we shipped 5 new features in 5 days, each shaped by real feedback from teams building voice agents.

I was excited about it, but I didn’t expect the results to be this amazing:

→ Reached ~125,000 people across channels
→ 250+ applications to our startup program – vapi.sh/Dm1wXZP
→ Dozens of thoughtful messages from teams already using what we shipped

This kind of feedback helps us move fast without breaking the wrong things.

If you’re working on voice agents, keep sharing what you’re running into. We’re listening.

You can always reach us at support@vapi.ai

Check out our blog for details on the latest features: vapi.ai/blog

When we first built Vapi, I had to airdrop contact cards to random people on the bus to get them to call it.

Last week, the Vapi team stood in front of 2,000 builders and watched someone lose their mind calling 1-844-HEY-VAPI.

We sponsored the @Socraticainfo Symposium, an event that turned a hockey rink into a massive open house of things people actually want to build: microchips in your living room, AirPods powered by skin, or apples engineered to taste sweeter.

~200 people came to the @YCombinator lecture with @nikhilro_ and @agupta about starting companies in college, the extraordinary AI boom happening in silicon valley, and the humorous realities of startup life.

~40 stayed after to chat.

~10 had already built something with Vapi.

That’s when it hit: we’ve crossed the SF chasm. Vapi’s out there, being used by builders we’ve never met.

Grateful to the Waterloo community and the Symposium team for hosting something this genuine.

A reminder that even as models and infra get more complex, the reason we build stays simple: help others build, faster.

“Hi, sorry I missed your call, leave a message after the beep…” Voice agents can’t tell the difference between what’s human and what’s machine.

Either:
– They start talking to a machine - False positives
– Or hang up on a real person - False negatives

Both are expensive. Both break trust.

Starting today, Vapi makes voicemail detection smarter.

Now you can:
- Choose your model for voicemail detection ( @google or @OpenAI )
- Customize listen time + prompt behavior
- Cut down on wasted calls

Read more about it here: vapi.sh/ZNghyRU

Stay tuned for more updates!

RAG is broken, at least for many customer-facing voice AI use-cases. Voice agents often hallucinate and confidently deliver wrong responses.

The traditional chunking methods limit context and product broken answers. Our developer community wanted us to fix the accuracy problem and we listened.

Today, we're launching Vapi’s Query Tool—the most requested feature from our community.

Instead of chunking, Query Tool selects a document and injects the entire document context and runs inference with large context window models.

Agents now have the full picture before providing a response, from your verified data.

Check out more information about our query tool here:

You know the drill—stay tuned for tomorrow!

@YCombinator's support helped us get off the ground. Now it's our turn to pay forward.

We want to support the small but growing voice AI community.

Startups are the driving force behind it, and they need every advantage they can get.

That's why we're launching the Vapi Startups Program:

- 100,000 free minutes, all models included
- Dedicated community, with monthly check-ins and technical support
- Influence over our product roadmap

The program is open to startups founded in the last 5 years, funded between $100K and $5M, where voice is core to the product.

Stop worrying about your runway, just get out there and ship.

Apply here: vapi.sh/Dm1wXZP

Launch week has been exciting so far! More coming tomorrow, stay tuned.