Deepgram 收購 Daily，承諾 Pipecat 維持開源

Deepgram announced on October 29, 2026, that it is acquiring Daily, the company behind the open-source Pipecat voice agent framework, unveiling the deal at its inaugural Deepgram Speak event in San Francisco. The Daily team will join Deepgram, and Pipecat will remain fully open source and free to developers under its existing BSD V2 license.

Deepgram said Pipecat and Daily’s global real-time network infrastructure will underpin its new Voice Agent Platform alongside its own speech models, forming what it described as the first complete, open, production-grade platform for conversational voice AI. The company characterized Pipecat as the world’s most popular open-source voice agent framework and said Daily’s network connects end users to voice agents with ultra-low latency and high reliability.

Daily co-founder Kwindla Hultman Kramer joins Deepgram as Head of Open Source. “Keeping Pipecat open was fundamental to our decision to join Deepgram,” Kramer said in the acquisition announcement. Deepgram CEO and co-founder Scott Stephenson said Kramer and the Daily team understand what it takes to keep a live conversation working reliably, and that acquiring Daily brings that expertise together with Deepgram’s to tackle the whole voice system and give developers a faster path from prototype to production.

Open-Source Commitments and the Pipecat Foundation

Beyond keeping the framework free under the BSD V2 license, Deepgram said it will invest in Pipecat’s continued development and community while keeping all of Pipecat open source and vendor neutral, so developers can choose and combine models, tools, network transport, and APIs from different providers. The companies are also establishing an independent Pipecat Foundation together with many of the organizations that have contributed to and use the framework.

In a separate post on Daily’s blog, Kramer wrote that the foundation will be managed by a steering committee representing the entire realtime AI ecosystem, including companies that compete with Deepgram. He wrote that nothing will change for Daily’s customers and the Pipecat ecosystem: Deepgram will sell, support, and develop all of Daily’s current products and services, including WebRTC, telephony, and cross-platform libraries, and Pipecat will remain vendor neutral. Kramer also wrote that Deepgram is adopting Pipecat as the core of its new voice AI products and services.

Kramer wrote that Pipecat has more than 200 integrations and almost 400 contributors. He credited 18 engineers with building the framework, giving special thanks to Mark, who he said has driven Pipecat’s development since its launch, and thanked the investors who led Daily’s rounds — David Waxman at TenOneTen, Avidan Ross at Root, Jenny Lefcourt at Freestyle, Lachy Groom, and Roseanne and Renata at Renegade Partners — along with board members Lee Edwards and Roseanne Wincek.

The Companies’ Shared History

Kramer recounted that he and Nina founded Daily to build infrastructure for realtime audio and video, which meant writing optimized packet-routing software, designing and deploying specialized compute clusters in dozens of locations globally, and writing client libraries that let developers embed realtime audio and video into web, desktop, and native mobile applications. Daily and Deepgram went through Y Combinator together in 2016, he wrote, and Deepgram has powered transcription inside Daily’s audio-video APIs since 2019.

Daily began building realtime voice agents and AI-native software experiments in 2023, Kramer wrote, and open sourced all of Pipecat in 2024: the core orchestration framework, client-side SDKs for every platform and language, the RTVI messaging and events standard, the Pipecat Smart Turn native-audio turn-detection model, SmallWebRTC transport, and the Pipecat Flows context engineering library. He described Pipecat as the most widely used voice agents and multimodal AI framework, saying it handles problems such as switching between models, low-latency network transport, streaming data between models, turn detection, and interruption handling.

According to Deepgram’s announcement, Daily’s infrastructure is a global network of 75 points of presence engineered for the latency and reliability demands of live AI. Deepgram said its own speech models and real-time infrastructure power conversational voice AI in production for more than 5,000 organizations globally, and that more than 650,000 developers have used Deepgram to process more than 200,000 years of audio.

Real-Time Context Standard and Next Steps

At Speak, Stephenson also introduced the Real-Time Context Standard (RCS), an open protocol co-developed with the Pipecat team. According to Deepgram, RCS addresses a gap in current voice AI systems: words pass from one model to the next, while the audio context around them, such as tone, who is speaking, and what is happening in the background, gets lost along the way. RCS defines three new types of metadata: context from the audio, such as a whisper or an unclear word; instructions for how a language model should use those signals; and guidance for how the response should sound. Adapters translate those signals and delivery instructions for each model, so developers can use the same approach with models from different providers.

The Real-Time Context Standard demo repository describes RCS as an open standard that carries that context in the same text stream, in formats any model in the chain can produce and read. Its demo voice agent handles customer service for an outdoor gear shop called Northwind Outfitters, chaining Deepgram Flux speech-to-text, a tone reader that reads each user turn for emotion and sarcasm, the PhoneLLM Mini language model, and ElevenLabs text-to-speech, with deployment on Pipecat Cloud. Deepgram invited developers, model providers, and teams building voice applications to test RCS in their own systems, share feedback, and contribute to the specification.

Later the same day at Speak, Chief Product Officer Ed Anuff was scheduled to preview the forthcoming Deepgram Voice Agent Platform and to introduce two new models built for live conversation, expanded language support, and a Global API designed to improve connection reliability and reduce latency. Deepgram said full technical details, demos, and availability for the day’s remaining product announcements would be published on its blog following Anuff’s keynote.