Deepgram Acquires Daily and Launches Protocol for Voice Agents

Deepgram announces the acquisition of Daily, creator of Pipecat, and commits to maintaining this framework as open-source. The company also launches an open protocol for voice agents, in a voice AI market marked by significant investments.

An Open Protocol and a Context Bus for More Expressive Agents

Deepgram is launching the Real-Time Context Standard, an open protocol for voice agents co-developed with the Pipecat team. This standard defines metadata such as audio context, instructions for language models on how to use these signals, and indications for the sound rendering of responses. Scott Stephenson states that this operation will allow developers to assemble end-to-end speech systems and agents with greater expressiveness, integrating different models through a context bus that connects everything. The agreement is also presented as a way to explore beyond voice, with Pipecat adapting to real-time communication in various formats. Kwindla Kramer emphasizes that Pipecat was never intended as a voice tool, but as a tool for a new type of AI-native software designed for real-time.

Acquisition Announced, Price Not Disclosed, and Open-Source Longevity of Pipecat

Deepgram announced on Thursday the acquisition of Daily, without disclosing the amount. Both companies clarify that Pipecat will remain open-source. Kwindla Kramer indicates that Scott Stephenson is committed to keeping it open-source and neutral towards suppliers in the long term, seeing a commercial interest in a rapidly growing ecosystem.

What Pipecat Does Today and What Deepgram Expects from It

Pipecat is a framework that manages audio and video streams and orchestrates voice transcription, large language models, and speech synthesis, with the goal of reducing latency. Its creator, Daily, operates a hosted infrastructure for all types of voice and video calls. The integration of Pipecat is presented as a way for Deepgram to deepen its understanding of real-time infrastructure. Scott Stephenson explains that he wants to open the ecosystem so that anyone can create voice agents and use the best models according to their needs, including partners if Deepgram's own models fall short, while also advancing them. He specifies that the idea of an orchestration framework for agents has been maturing for a long time, as part of a broader desire to work in open-source.

Crossed Paths and the Birth of Pipecat in the Era of GPT-4

Deepgram and Daily share common roots, having belonged to the same Y Combinator batch in 2016. Kwindla Kramer recounts that, during the release of GPT-4, Daily experimented with AI-human communication via voice and text, which led to Pipecat. According to him, the company received numerous expressions of interest for a buyout, but the commitment to preserving open-source weighed heavily in the choice of Deepgram, as did the alignment of engineering cultures that would give the team room to maneuver within a larger entity.

Recent Funding and Market Appetite for Voice AI

Daily has raised over $62 million, with investors such as Tiger Global, Highfield Capital, Nat Friedman, Lachy Groom, and Kevin Weil, according to PitchBook. Deepgram achieved unicorn status earlier this year thanks to a $130 million Series C round. More broadly, voice AI is attracting capital from both sides of the chain: model creators, such as ElevenLabs, Gradium, and Fish Audio, have raised millions, while infrastructure players like LiveKit and Vapi have secured multi-million dollar rounds this year.