Deepgram is moving up the voice AI stack. The San Francisco company, best known for speech recognition and synthesis models, announced on October 29 that it is acquiring Daily, a real-time audio and video infrastructure provider whose open-source Pipecat framework has become a common way to wire speech models, language models and telephony together into conversational agents. Neither side disclosed the price.
Pipecat is the strategic prize. Daily describes it as the most widely used framework for voice agents, with more than 200 integrations and close to 400 contributors, handling chores such as switching models mid-conversation, detecting when a speaker has finished a turn, coping with interruptions and streaming data between components. Daily open-sourced the full toolkit in 2024. Alongside it comes Daily's commercial network, which Deepgram says spans 75 points of presence built for low-latency live traffic.
Because Deepgram sells its own speech models, the obvious worry for Pipecat users is that the framework could tilt toward one vendor. Both companies addressed that directly. Pipecat stays free under its BSD license, and they are setting up an independent Pipecat Foundation governed by a steering committee that, according to Daily co-founder Kwindla Hultman Kramer, will include companies that compete with Deepgram. Kramer, who becomes Deepgram's head of open source, told TechCrunch that several suitors had approached Daily and that chief executive Scott Stephenson's promise to keep the project vendor-neutral settled the choice.
Stephenson framed the deal as a bet on open orchestration rather than lock-in. He told TechCrunch he wants developers to pick the best model for each job even when that is not a Deepgram model, and to treat gaps in Deepgram's own lineup as signals for where the company must improve. Deepgram plans to make Pipecat and Daily's network the base of a forthcoming Deepgram Voice Agent Platform, which its product chief Ed Anuff was due to preview at the company's first Speak conference in San Francisco, together with two new conversational models and a Global API aimed at more reliable connections.
The companies also published an open protocol, the Real-Time Context Standard (RCS), which Deepgram wrote together with the Pipecat team. Today, voice pipelines typically pass only transcribed words from one model to the next, discarding signals such as tone, who is talking or background noise. RCS adds three kinds of metadata: what the audio revealed, how a language model should act on those cues and how the spoken reply should sound. Adapters map the signals to each provider's models, so an agent could ask a caller to repeat an unclear word or lower its voice when someone whispers.
The two firms have history. They went through Y Combinator in the same 2016 batch, and Daily has used Deepgram transcription inside its audio and video APIs since 2019. TechCrunch, citing PitchBook, puts Daily's total funding above $62 million, while Deepgram reached a unicorn valuation earlier this year with a $130 million Series C.
The deal lands in a crowded market where infrastructure players such as LiveKit and Vapi and model makers such as ElevenLabs have all raised large rounds this year. Whether the foundation's governance gives rivals real influence, and how RCS fares if other model vendors decline to adopt it, will decide whether Pipecat stays the neutral layer Daily built.