Deepgram is acquiring Daily, the company behind Pipecat, one of the most popular open-source frameworks for voice AI infrastructure. The companies did not disclose the deal amount, but said Pipecat will stay open source. The move is part of a wave of consolidation in voice AI, where model makers and infrastructure providers are forming partnerships and acquisitions to strengthen their positions.

Pipecat is a framework that manages real-time voice and video streams together with pipeline components such as speech-to-text, large language models, and text-to-speech, while reducing latency. Daily, its maker, provides hosted infrastructure for voice and video calls. For Deepgram, the acquisition is expected to bring a deeper understanding of real-time voice and video infrastructure and an orchestration layer for building agents.

According to PitchBook data, Daily has raised more than $62 million from investors including Tiger Global, Highfield Capital, Nat Friedman, Lachy Groom, and Kevin Weil. Deepgram reached unicorn status earlier this year with a $130 million Series C. The two companies were in the same Y Combinator batch in 2016.

Daily co-founder Kwindla Kramer said the company started experimenting with AI-to-human communication over voice and text around the time GPT-4 came out, which led to Pipecat. He said there had been a lot of acquisition interest, but he favored Deepgram CEO Scott Stephenson’s commitment to keeping Pipecat open source forever, along with the matching engineering culture of the two organizations. “Scott was super committed to Pipecat being open-source and vendor-neutral forever,” he told TechCrunch, adding that there is both a personal commitment and a commercial driver: sitting in the middle of a fast-growing ecosystem, the best strategy is to keep everything as open as possible.

Stephenson said Deepgram has wanted to work in open source for a long time and saw an orchestration framework for agents as a potential area, which drove the decision to partner with Daily. He said Deepgram wants to open up the ecosystem so everyone can build voice agents, and that it is fine if developers use the best models for their use case rather than Deepgram‘s models. If Deepgram sees it needs to catch up in an area, it will work with partners and improve its own models and products. He also said the acquisition lets developers build agents with expressivity and end-to-end speech systems by plugging in different models and using a context bus that connects all parts.

Deepgram is also launching Real-Time Context Standard (RCS), an open protocol for voice agents co-developed with the Pipecat team. The protocol defines metadata such as context from the audio—for example, a whisper—plus instructions for a language model on how to use those signals and guidance for how the response should sound. Kramer emphasized that Pipecat was never thought of as a voice toolkit but as the toolkit needed to build AI-native software, and because the team came from real-time communication, they always assumed real-time. Deepgram said the deal lets it explore areas beyond voice, since Pipecat supports real-time communication in different formats.

Voice AI has seen heavy investment on both the model and infrastructure sides. Model makers such as ElevenLabs, Gradium, and Fish Audio have raised millions of dollars, while infrastructure companies like LiveKit and Vapi have raised multi-million-dollar rounds this year.