Deepgram to Buy Daily, Pipecat Gets a Foundation

Speech AI company Deepgram is acquiring Daily, the real-time video and audio infrastructure firm that created Pipecat, an open-source framework for building voice AI agents. Both companies announced the deal on 29 October. Neither has disclosed the price or said when the deal is expected to close.

As part of the deal, Pipecat will move to a new, independent Pipecat Foundation. Daily co-founder Kwindla Hultman Kramer wrote on Daily’s blog that the foundation will be run by “a steering committee that represents the entire realtime AI ecosystem, including companies that compete with Deepgram”. Deepgram’s announcement says Pipecat will keep its existing BSD licence.

The Daily team is joining Deepgram, and Kramer becomes Deepgram’s Head of Open Source. “Keeping Pipecat open was fundamental to our decision to join Deepgram,” he said in Deepgram’s release.

What changes for Daily customers and Pipecat users

According to Kramer, very little changes in the short term. “For Daily’s customers and the Pipecat ecosystem, nothing will change,” he wrote. “Deepgram will sell, support, and develop all of Daily’s current products and services. And Pipecat will remain open source and vendor neutral.”

Pipecat is a toolkit for wiring together the parts of a voice agent: speech recognition, a large language model, text-to-speech and the real-time network connection between them. Daily open-sourced it in 2024. The project now includes the core framework, client SDKs, the RTVI standard, a turn-detection model called Smart Turn, a lightweight transport called SmallWebRTC and a conversation-flow library, Pipecat Flows. Daily says Pipecat has more than 200 integrations and almost 400 contributors.

Daily also runs a global network of 75 points of presence for real-time audio and video, according to Deepgram. The two companies already have a long working relationship. Both went through the Y Combinator start-up programme in 2016, and Deepgram’s models have provided transcription inside Daily’s APIs since 2019.

TechCrunch, citing PitchBook data, reported that Daily had raised more than $62m. Neither company’s announcement gives a funding figure.

Deepgram’s plans: a full voice stack and a new context standard

Deepgram is best known for its speech-to-text and text-to-speech models. It says more than 5,000 organisations and more than 650,000 developers use its platform. The company describes the deal as a way to offer what it calls “the first complete, open, production-grade platform for conversational voice AI”. That is Deepgram’s own description, not an independent assessment.

Kramer wrote that “Deepgram is adopting Pipecat as the core of the company’s new voice AI products and services”. Chief executive and co-founder Scott Stephenson said: “Kwin and the Daily team deeply understand what it takes to keep a live conversation working reliably, and Pipecat gives developers the freedom to build with the models they choose.” He added that the acquisition would help Deepgram “tackle the whole voice system and give developers a faster path from prototype to production”.

Deepgram also announced a Real-Time Context Standard (RCS) at its Deepgram Speak event in San Francisco. It describes RCS as an open protocol co-developed with the Pipecat team. RCS defines three types of metadata that travel with a conversation: context drawn from the audio itself, such as a whispered phrase or an unclear word; instructions for how a language model should use those signals; and guidance on how the spoken response should sound. Adapters translate the signals for each model. “Our Real-Time Context Standard addresses what gets lost along the way,” Stephenson said, referring to the information that disappears when speech is turned into plain text. Chief product officer Ed Anuff is due to preview a Deepgram Voice Agent Platform, two new models, more languages and a Global API at the same event.

Why a foundation for Pipecat matters

The foundation is the most important part of this deal for anyone outside the two companies. Pipecat’s appeal has been its neutrality: developers can mix speech recognition from one vendor, a language model from another and a voice from a third. This month TSN has covered new speech models from Mistral, Rime and NVIDIA. All of them compete with Deepgram, and all are the kind of components a Pipecat developer might plug in. If Deepgram simply owned the framework outright, rivals would have a reason to doubt they would be treated fairly, and some developers might move elsewhere.

Moving Pipecat into a foundation with a steering committee that includes competitors is designed to answer that worry. The BSD licence also means anyone can fork the code if the project changes direction. Even so, the announcements leave open questions. Neither names the steering committee’s members, sets out how votes will work or explains how the foundation will be funded. Those details will decide whether “vendor neutral” holds up in practice, because Deepgram now employs the engineers who built the framework and plans to build its own products on top of it.

RCS raises a similar test. It is billed as an open protocol, but so far it comes from one vendor and one framework. It will matter only if other model makers add support for it.