This August, we focused on making agents easier to debug, safer in production, and faster to deploy. From a rebuilt logging stack to structured data capture and new models, here's everything we shipped.
New Features
Enhanced Logging & Observability
We rebuilt logging for clarity and speed. You can now watch live calls, trace every component, and export artifacts into your observability stack.
- Redesigned Call Logs UI: Real-time updates with precise filters.
- OpenTelemetry Exports: Send artifacts directly into your existing stack.
- Flexible Delivery: Deliver logs via webhooks or straight to S3.
- Component-Level Tracing: See STT, LLM, and TTS across a call.
- Live Debugging: Stream logs while conversations are in flight.
Open the Logging Dashboard →
Capture & Validate Data with Structured Outputs (Early Access)
Define a JSON schema and let the agent fill it during the call. The result is clean, predictable data that plugs straight into your APIs and databases.
- Schema Compliance: Guaranteed at every turn.
- Voice Forms: Collect structured inputs like intake or qualification.
- JSON Output: Ready for pipelines without extra parsing.
- Dashboard + API: Manage schemas and results flexibly.
- Model Choice: Pick the right extractor for your use case.
Structured Outputs details →
Request Early Access →
New from Vapi
Voice AI Weekly
A weekly brief for builders. We cut through the noise and highlight only what matters for your roadmap. Expect:
- Top 3 voice AI stories worth your attention.
- Spotlights on teams shipping with Vapi.
- A fast, scannable roundup of AI news.
Sign up for Voice AI Weekly. The next edition goes out this Thursday.
Sign up →
Platform Updates
🆕 New Models, Providers & Integrations
We expanded coverage, improved latency, and added more options across speech and language.
- Minimax Voice Provider: Global + China region support.
- GPT Realtime: Ultra-low latency voice streaming.
- GPT-5 Support: Via both OpenAI and Azure.
- Groq OSS Models: High-speed inference for open-source LLMs.
- Deepgram Nova 3: Stronger accuracy and broader languages.
- Gladia Solaria 1: Advanced multilingual transcription.
⚡ Developer Tools & API Enhancements
Sharper primitives for IVR, debugging, and safety.
- DTMF Keypad Input: More reliable IVR flows.
- PCAP Downloads: Pull packet captures for deep debugging.
- Security Filtering Plans: Configurable modes to protect sensitive data.
- Tool Rejection Handling: Better fallbacks and clearer error handling.
🚀 Platform & Infrastructure Upgrades
Operational improvements for smoother, more cost-efficient production.
- 60-Day Call Log Retention (Paid Add-On): Keep logs further back in time for audits, QA, or experiments.
- Optimized Workers: Better CPU and memory profiles.
- Smarter Autoscaling: Improved balancing under heavy load.
- Lambda Proxy with WebSockets: More reliable and faster transport.
🔊 Community & Events
VapiCon 2025
ICYMI, we added Olivia Moore (a16z), Michael Seibel (YC), and Justin Uberti (OpenAI) to the speaker lineup. To celebrate Labor Day, tickets are 30% off with code LABORDAY.
Get your ticket →
SF Voice AI Builders Meetup
Join us for the September edition on Friday, Sept 26 in San Francisco. Meet other builders, see live demos, and learn from teams shipping real systems.
Request to Join →
