Deepgram buys Pipecat maker Daily, keeps voice AI framework open source

Deepgram, a US company that sells speech-recognition and voice AI models to businesses, announced on Thursday, 29 October 2026 that it is acquiring Daily, the company behind Pipecat, a widely used open-source framework for building voice agents. The price was not disclosed. Both companies said Pipecat will stay open source and will be run by a new, independent Pipecat Foundation, while Daily's co-founder Kwindla Hultman Kramer joins Deepgram as its Head of Open Source.

Voice agents are AI systems that can hold a spoken conversation: they listen, work out what was said, decide on a reply using a large language model, and speak back. Businesses are deploying them in customer service, sales calls, appointment booking and help desks. Building one is harder than it looks. Several separate pieces must work together within fractions of a second, or the conversation feels slow and unnatural. Pipecat is a framework, a ready-made set of building blocks, that handles the audio and video streams and connects the speech-to-text, language-model and text-to-speech parts while keeping delays low. That is why many developers use it as the base of their voice products.

For Deepgram, owning the company behind that framework brings it closer to the developers who choose which speech models go into these systems. Analysts covering the deal have pointed out that Pipecat already ships with built-in support for Deepgram's speech-to-text and text-to-speech services, so the purchase may widen Deepgram's reach. At the same time, developers worry when a vendor buys a neutral open-source tool, because the tool can drift towards favouring the new owner's products. Deepgram's answer is the foundation structure: the companies say Pipecat will remain vendor-neutral, and rivals will sit on its steering committee. Kramer has said he chose Deepgram because its chief executive, Scott Stephenson, committed to keeping Pipecat open source.

The deal also shows where enterprise spending on AI is moving. After two years of chatbots that work through text, companies are now investing in systems that talk to customers on the phone and in apps. The infrastructure layer, meaning the plumbing that makes real-time voice possible, is becoming a battleground, and speech-model companies are buying tools to secure a place in it.

For Indian businesses, the shift is relevant because voice is how a large share of Indian customers prefer to deal with banks, telecom firms, insurers and online sellers, often in several languages. Indian start-ups and IT service firms building multilingual voice bots frequently rely on open-source frameworks such as Pipecat. Whether Pipecat stays genuinely neutral will matter to them when they choose which speech and language models to plug in.

Short-term: Developers using Pipecat should see no change in licence or access, according to both companies. Daily's existing customers will continue to be served by Deepgram. The new Real-Time Context Standard will be offered as an open protocol for others to adopt.

Long-term: Deepgram plans to combine Pipecat and Daily's real-time infrastructure with its own speech models into a single platform for building voice agents. If the foundation model works, it could become a template for how AI companies buy open-source projects without losing the trust of the developers who rely on them. If developers come to see Pipecat as tilted towards Deepgram, rivals may promote competing frameworks.

Who is affected: Developers and start-ups building voice agents; businesses that run customer-service and sales calls with AI; competing speech-model providers whose services plug into Pipecat; and IT service companies, including Indian firms, that build voice systems for clients.

Questions and Answers