Deepgram to acquire Daily, maker of Pipecat, and set up an independent foundation for the voice-agent framework

Speech AI company Deepgram announced on October 29 that it is acquiring Daily, the real-time audio and video infrastructure provider that created Pipecat, an open-source framework for building voice agents. The news came at Deepgram's first Speak event in San Francisco. Neither company disclosed the purchase price, the deal structure or when the transaction is expected to close.

What happens to Daily and Pipecat

According to both companies, the Daily team will join Deepgram, and Daily co-founder Kwindla Hultman Kramer becomes Deepgram's head of open source. Daily's post says Deepgram will continue to sell, support and develop all of Daily's existing products, including its WebRTC, telephony and cross-platform libraries and its global network, which Deepgram puts at 75 points of presence. Daily states that nothing changes for its customers.

Pipecat, which Daily open-sourced in 2024, will remain free under its existing BSD license. Deepgram says it is forming an independent Pipecat Foundation together with Daily and other organizations that use or contribute to the project. Daily writes that the foundation will be run by a steering committee meant to represent the wider real-time AI ecosystem, including companies that compete with Deepgram. Membership, funding and the committee's make-up have not been published. Daily says Pipecat counts more than 200 integrations and nearly 400 contributors.

The two firms have a long shared history: both went through Y Combinator in 2016, and Deepgram has supplied transcription for Daily's APIs since 2019.

An open protocol for conversational context

Alongside the deal, Deepgram published the Real-Time Context Standard (RCS), developed with the Pipecat team. In a post by chief product officer Ed Anuff, the company describes it as an extension of the chat-completions format used by most LLM APIs. It adds three kinds of metadata: signals about how the speaker sounded, instructions on how the model should treat those signals, and guidance on how the reply should be voiced by the text-to-speech system. Adapters are meant to translate these fields for different model providers. The specification is open for comment, with a demo on GitHub; no license is stated in the post.

Context and open questions

Deepgram says it is adopting Pipecat as the core of a forthcoming Voice Agent Platform, for which it has given no launch date. That makes the foundation's independence the main point to watch: the framework's commercial steward will now also sell a competing stack built on it. How the steering committee is chosen, and whether other speech-model vendors join, will indicate whether Pipecat stays as vendor-neutral as both companies promise.