Deepgram buys Pipecat maker Daily and keeps the framework open

A voice agent hands each turn through three models, one that writes down what the caller said, a language model that drafts the answer and one that reads it aloud, and Pipecat is the free code that runs that relay while keeping the gaps short. In its announcement, made at its first Speak event in San Francisco, Deepgram says Pipecat stays open source under its BSD licence, which lets anyone use and change the code for free, that an independent Pipecat Foundation will be set up with companies that use it, and that a forthcoming Deepgram Voice Agent Platform will bring its speech models and Pipecat’s open infrastructure together. Daily’s co-founder Kwindla Hultman Kramer becomes Deepgram’s head of open source; he told TechCrunch that other buyers had shown interest and that a promise to keep Pipecat open swung his decision.

Voice is already a paid line for marketers: ElevenLabs, one of the voice tools in our database, has paid plans from $6 a month for 30,000 credits with a commercial licence, on our 1 November check.

Deepgram’s release says Pipecat lets developers combine models from different suppliers, so a brand’s voice agent built on it can keep its current text-to-speech provider while moving transcription to Deepgram, and the open-source pledge is what keeps that mix available after the sale.