Skip to content
Neural Daily

BusinessNews

Deepgram buys Daily, the company behind Pipecat

🇭🇺 Olvasd el magyarul →

Voice AI infrastructure company Deepgram is acquiring Daily, maker of the popular open-source Pipecat voice agent framework. Pipecat stays open source under an independent foundation, and Deepgram also introduced a new open protocol, the Real-Time Context Standard.

By: Mikola Máté Patrik · · 3 min read

Graphic announcing that Daily and Pipecat are joining Deepgram on a purple background
Image: Deepgram · Press kit · Official social image from the Deepgram blog post · Source

What happened

Deepgram announced on October 29, at Deepgram Speak, its first conference, that it is acquiring Daily (Deepgram). Daily makes Pipecat, one of the most popular open-source frameworks for voice agents. Pipecat manages voice and video streams and connects the pieces of the pipeline, speech-to-text, the LLM and text-to-speech, with low latency. Daily also runs a global real-time network for voice and video calls.

  • The price was not disclosed.
  • Pipecat remains open source and free under its existing BSD license, and an independent Pipecat Foundation is being set up with contributing organizations.
  • The Daily team joins Deepgram.
  • Deepgram is building its new Voice Agent Platform on Pipecat and Daily's network, with timing to be announced.

TechCrunch reports that Daily has raised more than $62 million according to PitchBook, while Deepgram became a unicorn this year with a $130 million Series C (TechCrunch). Daily co-founder Kwindla Kramer said there was plenty of acquisition interest, but Deepgram committed to keeping Pipecat open source and vendor-neutral forever.

Real-Time Context Standard

Deepgram also introduced the Real-Time Context Standard (RCS), an open protocol co-developed with the Pipecat team. The problem it targets: in today's voice agents, only the words pass from one model to the next, while tone, speaker identity and background sounds get lost. RCS adds three fields to the chat completions format most LLM APIs already use (Deepgram):

  • how the person sounded, such as a whisper or an unclear word;
  • what the agent should do about it;
  • how the TTS model should speak the reply.

That lets an agent ask again when it didn't hear clearly, or answer quietly when someone whispers. Adapters translate the signals for each model, so the approach works across providers. Deepgram also announced two LLMs built for voice: Think-Live for the conversation itself and Think-Fast for the work behind it.

What it means for you

If you build voice agents on Pipecat, nothing changes in the short term: the license stays and the framework still works with any model. Longer term, Deepgram's money and team could speed up development, and the foundation is a safeguard against vendor lock-in. Keep an eye on RCS: if model providers adopt it, audio context could finally travel between STT, LLM and TTS in a standard way.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles