As Voice AI becomes more prevalent, applications need a way to store conversation data that supports observability, auditing, and long-term retention. The audio recording or a text transcript alone leaves out critical context. Additional details matter too, such as who was in the conversation, what devices were used, what was said, and how decisions were made.
vCons act as the “memory” of a conversation. They are structured, portable, machine-readable objects that capture conversation data across voice, video, messaging, and email, regardless of platform. This matters especially for Voice AI in regulated industries, like healthtech or fintech, where each conversation needs to be fully auditable.
Our guest is Jeff Pulver, VoIP industry pioneer and founder of the vCon Foundation. We cover the need for vCons, the current status of the IETF draft, where adoption stands, use cases and examples, and how AI is changing internet communications. We also preview the Fall ’26 Voice and Conversations on the Net Conference in Atlanta October 13-15.
The episode includes our monthly WebRTC Industry Update with Tsahi Levent-Levi. This month’s topic: Chrome’s Neural Echo Cancellation Leaves Edge, Electron, and Native Apps Behind.
WebRTC Live #118: Building the Data Layer for Voice AI with vCon
Episode highlights below.
Episode Highlights and Takeaways
A vCon is a JSON-based file format that puts structure around conversation data from calls, email, and text.
Attachments can add context to the conversation. Jeff said the structure lets a conversation leave one vendor’s system and load into another, or into an LLM, without training a model on each vendor’s storage format. He compared it to CSV for moving a spreadsheet between Google Sheets and Excel, and to a next generation of the call detail record. Until recently a vCon was defined as having at least one human party. That definition is being extended to agent-to-agent conversations.
A vCon gives an application memory that survives a dropped session and spans channels.
Jeff’s example was an airline chatbot that forces a customer to re-verify after a lost connection because it kept nothing from the first session. A vCon keeps that session and can be linked with related email, text, and Slack threads. In contact centers, he said, this lets a team analyze thousands of simultaneous calls for operational data no human could listen to at that volume. He has managed his own email with vCons for about nine months.
A vCon can record what each party consented to, including AI training, single use, and later withdrawal.
Jeff said a vCon can be signed and paired with SCITT (Supply Chain Integrity, Transparency and Trust) or other consent tooling. Content can be redacted while proof that the conversation happened remains. He said data theft will continue, and that vCon’s value lies in proving permission to place a call and to use the data. He predicted regulators will use vCon for enforcement within 12 to 24 months of the RFC.
Voice AI startups that already save conversations as JSON can adopt vCon with little work.
Jeff interviewed about 20 founders of recent AI startups over the summer, and none of the voice AI founders knew what vCon did. Several were already saving conversations as JSON and switched to saving them as vCons. He cited a 211 service in Kansas City, supported by the United Way, as a production deployment, and said messaging companies use vCons to manage consent.
Voice AI applications run on telecom infrastructure, and many of their builders have never worked in telecom.
Jeff expects the next disruptors in communications to come from that group. He listed the latency sources a voice AI app adds on top of network latency: text-to-speech, speech-to-speech, speech-to-text, and the LLM.
Jeff wants the vCon Foundation to work with regulators on disclosure practices for AI-initiated calls.
He said the FCC now treats AI-initiated calls as robocalls by default and that the EU applied similar rules as of August 2, 2026. The foundation is a 501(c)(6) and can lobby. He tied this to the Pulver Order, which the FCC issued on February 12, 2004 after he petitioned as an individual.
Jeff expects the vCon RFC within about six months and predicts 2027 will be “the year of vCon.”
His best guess is publication between mid-November 2026 and mid-March 2027. He expects interest to jump afterward, as it did when SIP was standardized. He also said agent-to-agent protocol work at the IETF, which he attributed to Jonathan Rosenberg, will build on vCon, and he pointed developers to the free hackathon at IETF 127 in San Francisco.
Jeff argued that vCon gives “unified communications” a concrete meaning: one record of how conversations developed across a company.
A new account owner can read ten years of account history, and a firm can recover what a departed engineer knew. He closed with personal use case examples, including saved voicemails from relatives and memory support for people with Alzheimer’s.
Up Next! WebRTC Live #119: WebRTC, MoQ, or Both? A QUIC Study
with Lorenzo Miniero, Chairman at Meetecho and author of the Janus WebRTC Server
Wednesday, November 4, 12:30 pm Eastern
