Deepgram buys Daily to add Pipecat to its voice AI stack

Deepgram is acquiring Daily, the company behind Pipecat, in a move aimed at strengthening its voice AI infrastructure, according to TechCrunch. The terms were not disclosed, but the companies said Pipecat will remain open source, a key point for developers building real-time voice agents.

For builders, the deal matters because voice AI products are increasingly stitched together from multiple moving parts: speech recognition, large language models, speech output, audio context and low-latency streaming. Deepgram is not just buying another tool; it is adding a framework that sits in the middle of those systems.

What Deepgram is getting with Pipecat

Pipecat is an open-source framework for real-time voice and video AI infrastructure. TechCrunch describes it as a system that manages voice and video streams, while also coordinating components such as speech-to-text, large language models and text-to-speech models.

That orchestration role is the practical center of the acquisition. Instead of treating a voice agent as a single model, Pipecat handles the pipeline around the model: live media streams, model handoffs and latency reduction.

Daily, Pipecat’s maker and maintainer, also provides hosted infrastructure for voice and video calls. Deepgram is best known for voice AI, and TechCrunch notes that Pipecat could give the company a deeper view into real-time voice and video systems.

The companies have a shared startup history. Both were in the same Y Combinator batch in 2016. Daily has raised more than $62 million to date, according to PitchBook data cited by TechCrunch, while Deepgram reached unicorn status earlier this year after a $130 million Series C round.

Why open source is central to the deal

The companies said they will keep Pipecat open source. That commitment is important because frameworks like Pipecat often sit between multiple vendors. If a developer is combining speech-to-text, a language model and a text-to-speech system, vendor neutrality can determine how flexible the final product is.

Daily co-founder Kwindla Kramer told TechCrunch that Daily had received acquisition interest, but Deepgram CEO Scott Stephenson’s commitment to keeping Pipecat open source and vendor neutral was a deciding factor. Stephenson also said Deepgram wants developers to use the best models for their use case, even when that means using partner models.

For creators, agencies and small software teams experimenting with voice agents, that matters in a concrete way. A framework that stays open can reduce the risk of building around a single closed stack too early. It may also make it easier to test different model combinations as speech systems improve.

Key pieces to watch after the acquisition:

• Whether Pipecat remains practical for non-Deepgram model combinations.
• How quickly Deepgram integrates Pipecat into its broader voice AI work.
• Whether the framework expands further beyond voice into real-time video and other formats.
• How developers respond to an open-source project moving inside a larger voice AI company.

Deepgram also introduces a voice-agent protocol

Alongside the acquisition, Deepgram is launching the Real-Time Context Standard, or RCS, an open protocol for voice agents that it co-developed with the Pipecat team.

According to TechCrunch, RCS defines metadata such as audio context, including signals like a whisper. It also includes instructions for how a language model can use those signals and guidance for how the response should sound.

That is a subtle but important layer for voice agents. In text chat, a model mainly reads words. In a real-time voice interaction, the system may need to interpret delivery, context and audio cues, then decide not just what to say but how to say it.

This fits a broader push toward lower-latency, more expressive voice systems. We have also covered related moves such as ElevenLabs releasing Eleven Realtime for faster speech-to-speech AI and Mistral AI launching Voxtral 3 for real-time voice agents. In the wider voice market, ElevenLabs is one of the model makers TechCrunch names as having raised significant funding.

What this means for voice AI builders

The acquisition shows how the voice AI market is starting to consolidate around infrastructure, not only models. TechCrunch points to funding on both sides of the sector: model companies such as ElevenLabs, Gradium and Fish Audio, and infrastructure companies such as LiveKit and Vapi.

For teams shipping voice AI products, the central question is becoming less about whether speech models can work and more about whether the full system can work reliably in real time. That includes managing audio streams, routing context, switching between models and keeping latency low enough for a natural exchange.

Deepgram’s bet is that owning more of that orchestration layer will help it serve developers building agents with richer speech systems. For Pipecat users, the near-term reassurance is that the project is supposed to stay open source. The longer-term test will be whether it remains broadly useful across the ecosystem as Deepgram builds more around it.