I'm excited to share that the Daily team is joining Deepgram. We will continue our work on the Pipecat open source framework and on our global network infrastructure that powers realtime audio, video, and AI applications.
We're joining forces with Deepgram because both companies believe every developer and enterprise will need models, realtime infrastructure, and open source orchestration, in every product.
For Daily’s customers and the Pipecat ecosystem, nothing will change. Deepgram will sell, support, and develop all of Daily’s current products and services. And Pipecat will remain open source and vendor neutral.
We are establishing an independent Pipecat Foundation. The foundation will be managed by a steering committee that represents the entire realtime AI ecosystem, including companies that compete with Deepgram.
Both Pipecat and our commercial infrastructure products — WebRTC, telephony, cross-platform libraries, and more — will benefit from the participation of Deepgram’s large, world-wide team of excellent engineers. Deepgram is adopting Pipecat as the core of the company’s new voice AI products and services.
When Nina and I founded Daily, our goal was to build the world’s best infrastructure for realtime audio and video. This meant writing optimized packet-routing software, designing and deploying specialized compute clusters in dozens of locations globally, and writing client libraries that allow developers to embed realtime audio and video into web, desktop, and native mobile applications.
Daily and Deepgram went through Y Combinator together in 2016. Some of my clearest memories of YC are the conversations we had with Scott, Deepgram’s CEO, about building large-scale, low-latency systems and about recent exciting advances in machine learning research. We first began partnering with Deepgram only a few years later, pre-generative AI. Since 2019, Deepgram has powered transcription inside our audio-video APIs.
Over time, both Deepgram and Daily grew to become trusted technology partners for some of the largest companies in the world. And both of our companies were in the right place at the right time to play key roles in the development of the incredibly exciting new wave of generative AI models, AI agents, and AI-native applications.
Deepgram’s focus on ultra low-latency transcription and voice generation enabled the development of conversational voice agents. Today, Deepgram’s customers deploy AI agents at production scale for customer support, reservations and scheduling, industrial embedded devices, education, robotics, and many other fast-growing use cases.
At Daily, we started building realtime voice agents and AI-native software experiments in 2023.
We quickly became convinced that our infrastructure and developer tooling, which we had created to power human-to-human communication, could also power human-to-AI communication.
As we experimented with human-to-AI communication, it became clear that software would change radically. In the future, most people, most of the time, will interact with most software via voice. Software will be re-imagined around voice as a primary input, plus voice, text, and rich, fluid, multi-modal, AI-generated output.
As we continued to build realtime AI agents and to embed LLM inference loops in more and more of our code, it also became clear that we needed new tooling to support a new kind of software. Pipecat is the toolkit that evolved as we both continued to experiment and began to help our customers deploy voice and video AI in production. Pipecat was so useful, and had so much potential across such a broad range of use cases, that the obvious next step was to make Pipecat open source.
Today, Pipecat is the most widely used voice agents and multimodal AI framework. It solves hard problems like easily switching between models, low-latency network transport, streaming data between models, turn detection, interruption handling, flexible context management, async tool calling, greedy inference, coordinating parallel inference loops, and standardizing interfaces for observability and memory and guardrails and much, much more.
We open sourced all of Pipecat in 2024: the core orchestration framework, client-side SDKs for every platform and language, the RTVI messaging/events standard, the Pipecat Smart Turn native-audio turn detection model, SmallWebRTC transport, and the Pipecat Flows context engineering library. Today, Pipecat has more than 200 integrations and almost 400 contributors.
Voice agents for customer support turned out to be the first “killer app” for voice AI. But people are building all kinds of software and systems with Pipecat. There are Pipecat-powered aerial drones and Pipecat toys for kids. Pipecat runs invisibly in some of the largest, most widely used applications in the world. You can build on-device and local/cloud hybrid AI applications with Pipecat on desktop, iOS and Android.
Pipecat is well on its way to becoming what one of our partners describes as “Linux for realtime AI.”
During this period, voice AI has accelerated — incredibly rapidly. Eighteen months ago most of our enterprise customer conversations were about how to build a first voice POC. Now, leaders in every industry understand how crucial it is to build voice AI customer touch points, and deploy them successfully. Our customers with "classic" human/human use cases are bringing AI features into their products or adding AI tools to their workflows.
Pipecat's growth over the past two years reflects that. We’ve worked with most of the companies training large language and audio models. We train and fine-tune open models, optimizing for low-latency and intelligence in long, multi-turn conversations. Increasingly, our team is interested in working on the hardest problems in every part of the new AI software stack. Today, our team does pioneering engineering work across all of these areas:
- Models
- Inference infrastructure
- Network infrastructure
- Agent hosting infrastructure
- Orchestration and tooling
- Applications
Today, we're joining Deepgram, in the belief that it's time for a new AI software stack —rooted in open source, and providing value at every layer of voice AI.
I'm really happy that a key part of this is the establishment of the Pipecat Foundation, where we're all committing to Pipecat's future as an open source standard.
And I'm thrilled I get to continue working on this stack alongside the engineers who built and believed in Pipecat as open source. Everyone in the voice AI ecosystem knows many of them: thank you Aleix, Brian, Chad, Christian, Dom, Filipi, James, Jon, Lewis, Marcus, Matt, Mattie Ruth, Paul, Rajneesh, Sunah, Vanessa, Varun. Special thanks to Mark, who has driven development of Pipecat since its launch. Eighteen engineers built the world's most widely used framework for this incredibly fast-growing sector of AI. (Believe in the power of open source!) For all of us at Daily, joining a larger company is an exciting step, allowing us to work deeply in every part of the stack, expand our WebRTC and realtime infrastructure, and bring Pipecat to even more developers.
Our goal is to create the standards and the standard toolkit for all software that has a human in the loop. We’re excited to join Deepgram to continue and to expand this vision.
Thank you also to the investors who believed in Daily and led each of our rounds: David Waxman at TenOneTen and Avidan Ross at Root, Jenny Lefcourt at Freestyle, Lachy Groom, Roseanne and Renata at Renegade Partners. Lee Edwards and Roseanne Wincek have been amazing partners on our board of directors. The startup ecosystem works because investors bet on big ideas early, and do hard work behind the scenes at every stage.
Never miss a story
Get the latest direct to your inbox.