Product Updates

We’ve shipped a series of backend updates focused on stability, developer experience, and smoother Replica handling:

• Stronger websocket reliability
• Vulkan performance tuning
• Cleaner default model naming and import handling
• Better operator and message signature support

You can now start conversations in test mode, where the replica does not join.

• Use cases: Validate setup, integrations, and flows
• Benefit: Test as much as you want without costs or concurrency usage

Docs updated here → Create Conversation docs

CVI now leverages chat history directly in the prediction process, giving your conversations better context continuity and more natural flow across turns.

This release also brings chat history integration, clears up duplicate imports, and updates the README for .env setup.

Search just got even smarter!

Our latest update enables fuzzy search functionality for personas, allowing users to find results based on partial matches of UUIDs or names, enhancing the search experience and improving user efficiency.

We’ve introduced stage direction support for replicas using ElevenLabs’ latest v3 alpha TTS engine.

This allows richer, more expressive conversations where replicas can adjust delivery based on embedded cues in the script.

‍

We’ve pushed a major optimization to Conversational Video Interface rendering.

• What’s new: Optimized zero-GPU-copy with a recycled buffer Vulkan renderer
• Impact: Up to 5–8 FPS boost (≈12% faster) depending on hardware
• Why it matters: Faster frame processing unlocks running two CVI instances on a single LS40S and frees ~5ms per frame for future perception models.

This is live in production now. Huge thanks to the engineering team for tackling tricky driver/OS challenges to make this possible.

‍

This week, we rolled out the next evolution in conversational video AI, and it’s all live in our portal.

🧠 Memories:

CVI now remembers you. Every conversation builds on the last, with full context, time/date awareness, and the ability to pick up exactly where you left off. Tutors can adapt to a student’s progress, therapists and mentors recall prior sessions.

Learn more about Memories >

📚 Knowledge Base (RAG):

Bring your own data, instantly. Upload docs or links and get lightning-fast, grounded answers in just 30ms, the fastest RAG on the market (15x faster than other RAG solutions.) Power AI recruiters, support agents, travel guides, and more.

Learn more about Knowledge Base >

🚧 Objectives & Guardrails:

Give your AI clear goals, branching logic, and measurable outcomes while keeping every conversation safe, compliant, and on-brand. Perfect for complex workflows or regulated industries.

Learn more about Objectives >

Learn more about Guardrails >

👤 Persona Builder:

No more wrestling with prompts. Just chat with our guided creation flow to shape AI personas with goals, behaviors, and style—and then test or launch within minutes. This makes spinning up an AI persona faster than ever.

Get started with Persona Builder >

✨ All four features work together to create highly capable, human-like AI agents that are fast, contextual, and trustworthy—redefining the way humans and AI interact like never before.

We’ve added an events console to the developer portal—making it easier to monitor everything happening during a conversation in real time.

From message flows to system activity, you now get a clearer view of how your AI Human is interacting under the hood.

You can now view full conversation details directly in the developer portal—including:

• 📄 Transcripts with speaker roles
• 👁️ Perception analysis

This update gives you clearer visibility into how your AI Human is seeing, hearing, and responding—all in one place.

If you're testing with memory, you might notice system messages in the transcript view. We're working on filtering those for a cleaner experience.

‍

A new update to the Tavus Developer Portal gives you more control over your AI personas.

🔧 What's new

• You can now enable/disable layers like Sparrow directly within a Persona
• Adjust sensitivity settings in real time
• Easily duplicate Personas from the side panel

Note: The current UX is a bit click-heavy and not super discoverable—but this is just the first step. A smoother, more intuitive flow is already in the works and should land in the next few days.

You’ll find the new buttons at the top of the right-side panel in the Portal.

‍