US Edition
Your source for latest news
TechnologyVoice AI

Voice-AI Rivals No More: Deepgram Buys Daily, Spins Pipecat Into a Neutral Foundation

The deal reunites two companies that trace back to the same Y Combinator class and creates an independent home for the open-source framework that much of the voice-agent industry has quietly standardized on.

PT
By PressTemps Technology DeskPublished Today, 09:06 ET · 6 min read
Voice-AI Rivals No More: Deepgram Buys Daily, Spins Pipecat Into a Neutral Foundation
A studio microphone on a boom arm, illustrative of the voice-recording and audio-AI infrastructure at the center of the Deepgram-Daily deal. Photo: Klim Musalimov / Unsplash
What to know
Deepgram, a venture-backed speech-AI company, announced on October 29 that it is acquiring Daily, the startup behind the open-source Pipecat voice-agent framework, with deal terms undisclosed.
Pipecat will be governed going forward by an independent, newly created Pipecat Foundation with a steering committee that explicitly includes Deepgram's competitors, and the code stays under its existing BSD 2-Clause open-source license.
Daily's co-founder and CEO Kwindla Hultman Kramer joins Deepgram as head of open source, and Deepgram says it will keep selling and supporting Daily's existing WebRTC, telephony and video products without interruption.
The companies also published a draft "Real-Time Context Standard" for voice AI pipelines and previewed a forthcoming Deepgram Voice Agent Platform built on Pipecat, though no launch date has been set.

Deepgram, a speech-recognition company based in San Francisco, said this week it is acquiring Daily, the real-time video and audio infrastructure startup behind the widely used open-source framework Pipecat. The companies announced the deal on October 29 at Deepgram's first in-person developer conference, Deepgram Speak, according to Daily's own announcement and a companion post on Deepgram's blog. Neither company disclosed the price or other financial terms of the transaction.

Under the agreement, Daily's entire team joins Deepgram, and Daily co-founder and chief executive Kwindla Hultman Kramer becomes Deepgram's head of open source. Deepgram says it will continue to sell, support and develop Daily's existing commercial products, including its WebRTC video and audio infrastructure, telephony connections and client libraries, so current Daily customers should see no immediate disruption to the services they already pay for.

Two companies, one lineage

The acquisition closes a loop that goes back further than most tech deals. Daily and Deepgram were both part of Y Combinator's 2016 cohort, and Deepgram has supplied the speech-to-text layer inside Daily's audio and video products since 2019, according to Kramer's post. Daily spent its early years building real-time video-calling infrastructure — the kind of plumbing that powers telehealth apps and virtual classrooms — before shifting toward voice-based AI agents starting in 2023, as large language models made it practical to build systems that could hold a spoken conversation rather than just transcribe one.

That pivot produced Pipecat, an open-source framework Daily released in 2024 for assembling the different pieces a voice agent needs — speech recognition, a language model, text-to-speech, and the real-time transport connecting them — into a single working pipeline. Deepgram describes Pipecat as the most widely used framework of its kind; Kramer's post puts it at more than 200 integrations and nearly 400 contributors, built by a team of 18 engineers since launch. The project ships under the permissive BSD 2-Clause license, meaning anyone can use, modify or resell software built on it without paying a licensing fee.

A foundation designed to reassure rivals

The structural wrinkle in the deal is what happens to Pipecat now that its creator belongs to a company that sells competing speech products. Deepgram and Daily say they are establishing an independent Pipecat Foundation, governed by a steering committee that is explicitly meant to include companies that compete with Deepgram, so that no single vendor controls the project's direction. Pipecat itself stays under its existing open license and, Deepgram says, will keep supporting speech and language models from other providers, not just its own.

"We're running toward open source, not away from it."

That line came from Ed Anuff, Deepgram's chief product officer, in the company's announcement post explaining why it chose to fold a competitor's open framework into its roadmap rather than build a closed alternative. Anuff wrote that some of Deepgram's longtime customers want to keep the option of mixing in other vendors' models, and that the company does not think they should have to give that up to get a production-ready product.

Alongside the acquisition, Deepgram published a draft of what it calls the Real-Time Context Standard, a proposed open specification for passing richer conversational signals — how a caller sounded, what the agent should do about it, and how a reply should be spoken — between the different components of a voice pipeline. The specification document, built on top of the chat-completions format most large language model APIs already use, is open for public comment, and Deepgram says it is inviting rival model and framework builders to weigh in.

Who the deal touches

The most immediate audience is the population of developers who have already built products on top of Pipecat — companies building customer-service bots, AI receptionists and other spoken-language applications that need to swap in different underlying models without rewriting their software. For them, the promise of an independent foundation is meant to address a familiar worry in open-source software: that a popular project will quietly become a sales funnel for whichever company ends up owning it.

Daily's existing commercial customers, who use its video and telephony infrastructure for applications unrelated to AI voice agents, are a second group with a direct stake in how smoothly the integration goes. Deepgram has committed publicly to keeping those products running and supported.

The deal also lands in a crowded and increasingly consolidating market. Deepgram raised $130 million in a Series C round in January at a $1.3 billion valuation and has been acquisitive, having previously bought the restaurant-voice startup OfOne and the meeting-transcription company Poised. Daily, for its part, had raised venture funding from backers including TenOneTen Capital, Root Ventures and Freestyle Capital since its 2016 founding, according to figures reported by trade outlets including audioXpress, though neither company disclosed the specific terms of the acquisition itself.

What comes next

Deepgram says the acquisition supports a forthcoming product it calls the Voice Agent Platform, which would combine Deepgram's speech models with Pipecat's orchestration layer and Daily's real-time network — Daily says it operates infrastructure in 75 points of presence worldwide — so customers could run voice agents either through Deepgram's hosted service or inside their own cloud environment. The company has not announced a launch date. In the meantime, Deepgram rolled out expanded language coverage for its existing products at the same event, including multilingual speech recognition in 55 languages and text-to-speech generally available in nine languages with dozens more in early access.

The Real-Time Context Standard's comment period, and the formal setup of the Pipecat Foundation's steering committee, are the two concrete items to watch in the coming weeks — both will determine whether competitors treat Deepgram's promise of neutrality as credible or decide to fork the project and go their own way.

More on this story

All Technology