Deepgram Acquires Daily, the Owner of Pipecat

Speech recognition and voice AI infrastructure company Deepgram has acquired Daily, the company behind the open-source Pipecat framework. The deal was announced on October 29, 2026 at the first Deepgram Speak event in San Francisco. The Daily team will join Deepgram, while Pipecat will remain fully open source and free for developers under its existing BSD V2 license.

Pipecat will remain open, independent foundation being established

Deepgram said it will invest in Pipecat’s development and community, and that the framework will remain vendor-neutral. The goal is to let developers freely choose and combine models, tools, network transport layers, and APIs from different providers.

The companies are also establishing an independent Pipecat Foundation together with organizations that contribute to Pipecat. Daily co-founder Kwindla Hultman Kramer wrote that the foundation will be governed by a steering committee that includes Deepgram’s competitors and will represent the entire real-time AI ecosystem. Kramer will serve as Head of Open Source at Deepgram.

According to Kramer, Pipecat has more than 200 integrations and nearly 400 contributors. In 2024, the framework was open-sourced along with its core orchestration layer, client SDKs for all platforms, the RTVI messaging standard, the turn detection model Smart Turn, the SmallWebRTC transport layer, and the Pipecat Flows context library. There will also be no changes for Daily’s existing customers; WebRTC, telephony, and cross-platform libraries will continue to be sold and supported by Deepgram.

Voice Agent Platform and a 75-point network

Deepgram is building the foundation for its new Voice Agent Platform by combining Pipecat and Daily’s real-time network infrastructure with its own speech models. Daily’s 75-point global network, designed for live AI traffic, promises low latency and high reliability.

According to the company, speech-based AI systems are currently running in production at more than 5,000 organizations, and more than 650,000 developers have used Deepgram’s infrastructure. The two companies also share a long history: Deepgram and Daily were part of the same Y Combinator batch in 2016, and Deepgram has provided transcription for Daily’s audio-video APIs since 2019.

Real-Time Context Standard

Another innovation introduced at the event was the open Real-Time Context Standard (RCS), developed together with the Pipecat team. According to Deepgram, in today’s voice AI systems, context around the audio is lost as words pass from one model to another; who is speaking, whispering, unintelligible words, and background sounds cannot be carried through the chain.

RCS defines three new types of metadata: context from the audio, instructions for how the language model should use these signals, and guidance on how the response should sound. Because adapters translate these signals for each model, the same approach can be used with models from different providers. Deepgram invited developers, model providers, and teams building voice applications to try the standard in their own systems.

The company also introduced a preview of the Deepgram Voice Agent Platform and two new models optimized for live conversation on the same day. The move to bring an open-source framework and commercial infrastructure under one roof in the voice agent market aims to lower developers’ cost of moving from prototype to production.

Source: Unite.AI