On October 29, 2026, the speech model company Deepgram announced that it is acquiring Daily, which runs realtime audio and video infrastructure and maintains Pipecat, an open-source framework for building voice and multimodal agents. The price was not disclosed.
What both companies commit to
Deepgram's release says Pipecat "will remain fully open source and free to all developers under its existing BSD V2 license." The GitHub API lists the repository's licence as BSD-2-Clause. Daily's co-founder writes that the companies are "establishing an independent Pipecat Foundation", to be "managed by a steering committee that represents the entire realtime AI ecosystem, including companies that compete with Deepgram." He joins Deepgram as Head of Open Source.
Daily's post also says "Deepgram is adopting Pipecat as the core of the company's new voice AI products and services." Deepgram plans a Voice Agent Platform that combines its speech models with Pipecat and Daily's network. That is the tension to watch: the framework's largest corporate user will now employ most of its maintainers.
What Pipecat does
A voice agent chains a speech-to-text model, a language model and a text-to-speech model over a live connection. Daily's post lists the hard parts Pipecat handles: switching between models, low-latency transport, turn detection, interruption handling, context management and async tool calls. It says the project has more than 200 integrations and almost 400 contributors. The repository had about 17,000 stars on October 31, which measures attention, not quality.
A protocol for what text loses
Deepgram also introduced the Real-Time Context Standard (RCS), an open protocol co-developed with the Pipecat team. The release describes the gap it targets: "the words get passed from one model to the next, but the rich audio context around them, like tone, who's speaking and what's happening in the background, gets lost along the way."
RCS defines three kinds of metadata: context from the audio, such as a whisper or an unclear word; instructions for how the language model should use those signals; and guidance for how the reply should sound. Adapters translate them for each model provider. The demo repository was created on October 28 and had no licence file when we checked on October 31.
What remains uncertain. The foundation's charter, membership and trademark ownership have not been published. "Most widely used" is the companies' own description. RCS has one sponsor so far and no independent implementation yet.
Analysis: When a company buys the maintainers of an open-source project you depend on, check three things: the licence, which here stays permissive; who controls the repository and the name; and whether the release roadmap starts to favour the buyer's own models. For Belgavi's Book Store, a phone line where Geetha's customers ask about a CASE-3104 return could be built on Pipecat with any speech vendor. Keep the model adapters behind your own interface so that switching vendors remains a configuration change. Treat RCS as an experiment until a second vendor implements it.
Sources and evidence
Deepgram is acquiring Daily, maintainer of the Pipecat voice agent framework (announced October 29); both firms say Pipecat stays BSD-2-Clause and will be run by an independent foundation whose steering committee includes Deepgram competitors, while Deepgram builds its own platform on it and introduced the Real-Time Context Standard for carrying audio context between models. The foundation's charter and governance are not yet published.