Deepgram buys Pipecat maker Daily and keeps the voice AI framework open

Voice AI company Deepgram is buying Daily, the maker of Pipecat, an open-source framework for building voice agents. Pipecat will stay open source and free under its existing BSD V2 licence. The companies are also setting up an independent Pipecat Foundation.

Deepgram announced the deal on Thursday in a blog post, on the day of Deepgram Speak, its first conference, in San Francisco. Daily co-founder Kwindla Hultman Kramer set out the details in a post of his own. Neither company disclosed the price of the Deepgram Daily acquisition, Ivan Mehta reported for TechCrunch.

“Keeping Pipecat open was fundamental to our decision to join Deepgram,” Kramer said in the announcement.

What Deepgram is buying

Pipecat links the parts of a voice agent: speech-to-text, a large language model and text-to-speech. It also handles the voice and video streams between them while keeping delays low. Kramer wrote that it has more than 200 integrations and almost 400 contributors, and that 18 engineers built it.

Daily runs hosted infrastructure for real-time voice and video on a network of 75 points of presence, according to Deepgram. Daily has raised more than $62m from investors including Tiger Global, Nat Friedman and Lachy Groom, TechCrunch reported, citing PitchBook data.

Deepgram says more than 5,000 organisations run its speech models in production. It raised a $130m Series C in January that made it a unicorn, according to TechCrunch. Both companies went through Y Combinator in 2016, and Deepgram has powered transcription inside Daily’s products since 2019, Kramer wrote.

A foundation for Pipecat

The Daily team will join Deepgram, and Kramer becomes its head of open source. Deepgram will sell, support and develop all of Daily’s current products, he wrote. It is also making Pipecat the core of its new voice AI products.

“For Daily’s customers and the Pipecat ecosystem, nothing will change,” Kramer wrote.

A steering committee will manage the foundation, and it will include companies that compete with Deepgram, according to Kramer. He told TechCrunch that several companies had shown interest in buying Daily. He chose Deepgram because chief executive Scott Stephenson committed to keeping Pipecat open source.

“Scott was super committed to Pipecat being open source and vendor-neutral forever,” Kramer told TechCrunch.

Stephenson told TechCrunch that Deepgram had wanted to work in open source for a long time. He said developers should be free to pick models from other companies.

“It’s great if you’re using Deepgram models, but I want you to be using the best models for your use case,” Stephenson said.

A standard for what voice agents hear

At Speak, Stephenson is also introducing the Real-Time Context Standard (RCS), an open protocol built with the Pipecat team. Deepgram says voice systems today pass words from one model to the next. What gets lost is the audio around them, such as tone, who is speaking and background noise.

RCS adds three kinds of metadata. The first is signals from the audio, such as a whisper or an unclear word. The others are instructions for how a language model should use them, and guidance on how the reply should sound. Deepgram says that lets an agent ask again when it did not hear clearly, or answer quietly when someone whispers.

Chief product officer Ed Anuff is due to preview the Deepgram Voice Agent Platform, built on Deepgram’s models and Pipecat, later on Thursday. He will also introduce two new models for live conversation, the company says.

Voice AI has drawn large investments this year from both model makers and infrastructure companies, TechCrunch noted. TNW has covered Vapi’s Series B, Fish Audio’s seed round and talks over an ElevenLabs tender offer.