Voice AI is consolidating, and Deepgram just moved. The speech recognition vendor said Thursday it is buying Daily, which maintains Pipecat, among the most widely used open source voice agent frameworks. No price was disclosed; both sides pledged Pipecat stays open source.
Pipecat handles the unglamorous middle of a voice pipeline, shuttling audio and video between speech-to-text engines, large language models and text-to-speech synthesis while holding latency down. Outside Pipecat, Daily runs hosted plumbing for live calls. Owning that layer puts Deepgram at the stack’s center instead of one end, and reaches beyond voice, since the framework serves real-time communication in other formats.
Chief executive Scott Stephenson said the firm had long wanted to work in open source and picked agent orchestration as the way in. Developers, he argues, should plug in the strongest model for their use case, even a rival’s. The deal lets builders assemble agents by swapping models through a context bus linking every component.
Daily co-founder Kwindla Kramer said buyers had shown interest, but he chose Deepgram on Stephenson’s permanent, vendor-neutral commitment to Pipecat and a matching engineering culture. Around the GPT-4 era, Daily began testing AI-to-human exchange by voice and text, work that became the framework. Kramer never saw Pipecat as a voice toolkit alone; it was what building AI-native software required, real-time by default.
PitchBook puts Daily’s earlier haul above $62M, from Tiger Global, Highfield Capital, Nat Friedman, Lachy Groom and Kevin Weil. Deepgram hit unicorn status this year on a $130M Series C, and the two firms share a 2016 Y Combinator origin.
Deepgram also is launching the Real-Time Context Standard, an open protocol built with the Pipecat team that defines metadata like a whisper in the audio, how a model uses those signals and how the reply should sound.
