January Updates 2: AI-Powered Summaries, Observations & Open-source Updates

This January update marks a pivotal moment for rtcStats. As our second release of the year, we are officially moving to Version 1.0.

This transition signifies our evolution beyond the "early stage" service phase into a mature, production-ready platform. The catalyst for this milestone is the integration of our sophisticated AI engine, which transforms raw WebRTC data into actionable narratives.

🚀 Major Improvements

AI Summary: Your Call, Narrated ✨

Our biggest update in version 1.0 is the AI Summary engine: Instead of manually parsing complex logs, our AI digs into the rtcStats engine to extract a high-level overview from three critical perspectives:

AI Summary: Your Call, Narrated

• Network: Identifies congestion, packet loss, and latency trends.
• User: Highlights behavioral patterns like hardware changes or frequent muting.
• Media: Evaluates codec performance and bitrate stability.

Note: This feature requires a Developer or Enterpriseplan. It runs automatically in the background once you upload a file, allowing you to understand a call's health at a single glance.

Just so we're clear - this is our first step into AI. We've got more lined up in our roadmap for 2026 😉

New Observations 🖥️

We've expanded our "Observations" engine to detect more nuanced edge cases that can degrade user experience:

New Observations

• Too many getUserMedia requests: Catches logic loops in the application.
• Excessive mic hardware changes: Detects bad behavior that can ruin the experience.
• Turn Error 701: Specifically flags failed relay connection establishment.
• SDP Analysis: Identifies less efficient simulcast detection and redundant lines in the Session Description Protocol.

🛠️ Other Enhancements

Server & Stack Detection 🔍

To give you better context on how calls are being routed, rtcStats now automatically identifies the underlying infrastructure. We've added detection for:

• Media Servers: Janus, Asterisk, Freeswitch, and Pion.
• Stacks: PJSIP, Pixel Streaming (Unreal Engine), and ICE-Lite.

These are now surfaced directly in the Session Details, and this context is used for the AI Summary.

UI & Quality of Life

We've polished the interface to handle large data more gracefully:

• Better Scrolling: Navigating large device lists is now seamless.
• Improved Tables: The streams and data-channels tables have been redesigned for better alignment and readability.
• Empty States: Clearer feedback when no dumps are available or filters return no results.

🐞 Bug Fixes

Accuracy is the foundation of debugging. In this release, we have focused on refining our core computation engines:

• RTT & MOS Computations: We fixed a bug where missing values in stats led to skewed RTT (Round Trip Time) and MOS results.
• UserAgent Handling: We resolved issues with userAgentData to ensure that browser and OS identification is consistent across all reporting modules especially when using the new rtcStats format.
• Stability: Fixed several crashes related to the latest Next.js updates and handled edge cases where peer connections were missing.

🔧 Open Source - rtcStats

Our commitment to the community continues with a new release of our client-side library.

rtcStats-js (2.1.0)

The latest version of rtcStats-js is now available. This update specifically targets an issue with bad timestamps that occurred in certain browser environments. Upgrading to 2.1.0 ensures that your data collection remains perfectly synchronized, providing a reliable foundation for the AI Summary engine.

Happy debugging with rtcStats and its new AI superpowers! 💎