Griffin, Face Studio, Connectors, Starter Kits, and a live PAL Maker webinar on Nov 5.  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ 

Tavus October 2026 Recap

Hey [First name],

It's been the biggest month in Tavus history. If you're new here: Tavus builds PALs, human-like AI that can see you, hear you, and speak to you like a person would. More than 200,000 people build on Tavus, and we are excited to see what you create next.

This month we previewed Griffin, the first Human Interaction Model, a new class of model built to understand and generate face-to-face conversation in real time. It listens while you talk, reads your expressions and your pauses, and renders every frame live, all in one model. Within a few hours the launch post had passed 21 million views on X, and dozens of outlets covered the future of human computing.

We also shipped Face Studio, MCP Connectors, Starter Kits, an official ChatGPT plugin, support for GPT Live 1.5 and Agora, and a long list of smaller upgrades that make PALs faster and easier to build. And on November 5 we're hosting a live PAL Maker webinar. Here's the rundown:


🎁 3x minutes for 3 months

Upgrade to any paid plan and we'll triple your included minutes for your first three months! Upgrade in the portal and use code 3XTAVUS


Griffin: the first model to pass the video Turing test face to face

Griffin is the world's first Human Interaction Model (HIM), a new class of model designed to understand and generate face-to-face real-time human interaction. Griffin combines our earlier research into real-time perception, conversation and expressive human-like video into one unified video to video system.

Everything we've built so far, Phoenix for the face, Sparrow for turn-taking, Raven for perception, has been working toward one goal: talking to a machine should feel as natural as talking to a friend. What makes that hard is all the stuff you never consciously notice in a good conversation: the little expressions, the way someone shifts in their chair, knowing when to jump in and when to wait. Griffin is the first model that gets those right, because it does all of it at once instead of handing off between separate systems.

  • 48% thought it was a real person: we put Griffin on a one-minute video call with people who were told they'd be talking to another participant, then asked whether their partner was real. Previous systems, including our own, convinced under 3% of participants. With Griffin, almost half never considered that it might be AI.
  • #1 on NVIDIA's benchmark: NVIDIA ran its own independent evaluation on VideoFDB, its benchmark for face-to-face AI, and Griffin came out #1 on both tracks. That puts it ahead of Gemini and OpenAI's realtime model, and 37% ahead of the next best AI at reacting in the moment.
  • One model, not a pipeline: every AI you've talked to before was really a chain of separate models. Speech recognition hands off to a language model, which hands off to a voice, which hands off to a face, and each one waits for the last. Griffin does all of it in one model. It listens while it talks, decides when to speak and when to hold back, and generates every pixel of the frame live from a single photo. Pause to think and it waits for you. Interrupt it and it actually stops talking.

To understand its capabilities, watch the demos below to see it in action. In one, Mars plays Simon Says with it. In another, Sagar solders a circuit board while it watches over his shoulder and talks him through it.

Watch it in conversation · Read the research

Where Griffin stands

A model that can pass the Turing test can also convince someone it isn't AI, and we take that seriously. Griffin-Lite is a research preview for a small group of trusted testers while we finish disclosure features and work with AI safety organizations on evaluations. We expect Griffin to be available for existing power users in the near future. If you'd like to be part of the testing, the waitlist is the place to start.


Customer spotlight: Elevare, a care companion for new moms

"Sometimes when we say things, we express the opposite. We're not saying what our words are saying."

That's Kayla Walsh, founder of Elevare Health. She grew up watching untreated postpartum depression take over her mother's life, and now dedicates her own life to supporting new moms. Most postpartum care ends at the six-week checkup, but nearly 38% of pregnancy-related deaths happen between six weeks and a year after birth.

So Elevare built a support system that a mom can talk to face to face at any hour, from pregnancy through her baby's first year, while her care team makes every clinical decision. It checks in between visits, and it's built to notice when a mom's words and her tone don't match. Before choosing Tavus, Elevare tested several vendors. The companion had to keep listening through a crying baby, read emotion in both directions, and never leave a pause long enough to feel like judgment. "That's something none of the other vendors we evaluated do well."

Read the story · Watch the video


Connectors

Connect Linear, Gmail, Google Calendar, Notion and twenty plus apps and add them to your PAL. Whatever you'd open those apps to do, ask your PAL instead and watch the magic happen. Your PAL hands the job to a background agent and keeps the conversation going. A short status line under its face shows what it's working on, and it tells you when the job lands.

Add a connector in PAL Maker

What your PAL can do now

  • Catch up. Ask what you missed in your inbox. It reads the threads, checks the open issues in Linear and gives you the short version.
  • Close out a meeting. Ask what you agreed yesterday. It reads the Granola notes back, files the action items in Linear, posts the summary to Notion and books the follow-up on Google Calendar.
  • File the bug. A customer says the export button just spins. It finds their session in LogRocket, files the issue in Linear with the replay attached and emails the engineers.
  • Ship the page. Describe a waitlist page. Webflow builds and publishes it, Stripe wires the checkout, and you're still on the call when the link goes live.
  • Pull the numbers. Ask how the funnel did this week. It reads the conversion from Amplitude and the revenue from Stripe, and tells you out loud.

What else shipped

  • Face Studio is live in PAL Maker. Upload a photo, or describe the face you want: "a cheerful museum guide in her sixties with silver curls and reading glasses." A preview is ready in about two minutes.
  • Starter Kits are ready-made PALs for real work: onboarding new hires, interviewing candidates, coaching customer service reps, running compliance training. Pick one, make it yours, and go live in a few clicks.
  • Build PALs inside ChatGPT. Our official ChatGPT plugin runs on the Tavus MCP server with Charlie behind it. The same server also connects to Claude Code, Cursor, and Codex.
  • Charlie now has the internet. He pulls from websites and looks up recent information while he builds your PAL. Nothing to switch on.
  • Building with v0, Lovable, Replit, or Bolt? We shipped better agent instructions alongside the Tavus MCP, so those tools build PALs the right way on the first try.
  • GPT Live 1.5: put a Tavus face on OpenAI's GPT Live 1.5 speech-to-speech model through Pipecat or LiveKit. Docs
  • Agora: Tavus now officially supports Agora, so you can deploy PALs inside its low-latency, real-time WebRTC framework. Docs
  • Pipecat and LiveKit: lower audio-to-frame latency and a handful of other pipeline optimizations, so conversations feel smoother on both frameworks. Pipecat docs · LiveKit docs

Live webinar: Getting Started with Tavus PAL Maker

Thursday, November 5 at 10:00 AM PT. New to Tavus, or haven't built in a while? Join us live as Sagar, our head of product, builds a PAL from scratch: a custom face and voice, the personality, connectors, testing, and deploying it. Bring your questions and we'll answer them as we go. Can't make it live? Register anyway and we'll send you the recording.


Customer spotlight: Ruby, a face-to-face AI for doctors

Med Learning Group, a medical education company, built Ruby for the doctors who treat skin cancer. A doctor can talk to her at any hour about a hard case, every answer comes only from research their field trusts, and the doctor makes the call. One oncologist, after a session: "The thing that surprised me is that she's so accurate, captures what I said, and she can adapt to my question." Watch Ruby in action

As always, reach out with any questions or ideas you have! We're happy to help.

Faster faster,
The Tavus Team


You're getting this because you have a Tavus account.

Tavus Inc., 2261 Market Street STE 86391, San Francisco, CA 94114

Unsubscribe · Manage preferences