Deepgram is acquiring Daily, the company behind Pipecat, and Pipecat will stay open source. Deepgram announced the acquisition on October 29, 2026, at its first Deepgram Speak event in San Francisco, and said the framework will remain fully open source and free under its existing BSD 2-Clause license. The companies are also creating an independent Pipecat Foundation, run by a steering committee drawn from across the realtime AI ecosystem, including companies that compete with Deepgram (Daily's announcement, published October 29, 2026, retrieved 2026-10-30). For developers building voice agents, the headline is not the deal itself. It is the structure around it: a popular open source orchestration layer, a commercial realtime network, and a foundation meant to keep the first of those two from becoming a single vendor's tool.
Key Takeaways
Deepgram announced on October 29, 2026 that it is acquiring Daily, creators of Pipecat, the open source framework for realtime voice and multimodal AI agents.
Pipecat will remain fully open source and free under its existing BSD 2-Clause license, and Deepgram says it will stay vendor neutral, so developers can keep mixing models, tools, transports, and APIs from different providers.
The companies are establishing an independent Pipecat Foundation, managed by a steering committee that represents the whole realtime AI ecosystem, including Deepgram's competitors.
Daily's team joins Deepgram, and Daily co-founder Kwindla Hultman Kramer becomes Deepgram's Head of Open Source. Deepgram will also sell, support, and develop Daily's current products, including WebRTC and telephony.
Alongside the deal, Deepgram introduced the Real-Time Context Standard, an open protocol that carries audio context, such as tone and unclear words, between models instead of passing only transcribed words.
Acquisitions of open source projects usually raise one question first: what happens to the code? Here the companies answered that question before anything else, in the first paragraphs of both announcements. That does not settle every question. The foundation's committee members and setup date have not been named yet, and the companies did not disclose a purchase price (TokenPost's report). But the commitments are unusually specific for a deal announcement, and they are worth reading closely, because they define what developers should hold Deepgram to in the months ahead.
What Is Pipecat?
Pipecat is an open source Python framework for building realtime voice and multimodal conversational agents. That is how its GitHub repository describes it, and the description matters because Pipecat is not a speech model. It is the layer that connects the pieces of a voice agent. Those pieces are speech recognition, a language model, and speech synthesis. They also include the transport that carries audio in realtime, the tools the agent can call, and the conversation state in between. A single voice agent, or a system of specialist agents that hand work to each other, can run on one shared pipeline.
Daily began building realtime voice agents and AI experiments in 2023, after years of building infrastructure for human to human audio and video calls. Pipecat grew out of that work. Daily open sourced all of it in 2024. The release included the core orchestration framework, client software for every major platform, the RTVI messaging standard, the Pipecat Smart Turn turn detection model, the SmallWebRTC transport, and the Pipecat Flows library for structured conversations (Daily's announcement). Today Pipecat has more than 200 integrations and almost 400 contributors, according to the same announcement. Its repository, retrieved 2026-10-30, shows 16,923 stars, 2,998 forks, 14,728 commits, and 125 releases under the BSD 2-Clause license (pipecat-ai/pipecat, GitHub, retrieved 2026-10-30).
The problems Pipecat claims to solve explain why it spread. Switching between models without rewriting the agent is hard. Moving audio with low latency is hard. Streaming data between models, detecting when a speaker has finished a turn, handling interruptions, managing context, calling tools asynchronously, and coordinating parallel work are also hard. Together, they are harder than apart. Daily co-founder Kwindla Hultman Kramer writes that Pipecat runs in customer support, reservations and scheduling, industrial devices, education, and robotics. It also runs in aerial drones and children's toys. He writes that it runs invisibly inside some of the largest applications in the world (Daily's announcement). The claim that Pipecat is becoming, in one partner's phrase quoted by Kramer, the Linux for realtime AI, is marketing. The breadth of use cases is not: a framework that only handled phone calls would not show up in drones and toys.
Aspect | Deepgram acquires Daily |
|---|---|
Announced | October 29, 2026, at Deepgram Speak in San Francisco |
Buyer | Deepgram, voice AI infrastructure company |
Acquired | Daily, creators of Pipecat and a global realtime network |
Pipecat license | BSD 2-Clause, unchanged and free |
Governance | New independent Pipecat Foundation with an ecosystem steering committee |
Daily team | Joins Deepgram; Kramer becomes Head of Open Source |
Price | Not disclosed |
Also announced | Real-Time Context Standard, an open protocol |
What Deepgram Is Actually Buying
The deal has three parts, and only one of them is code. The first is Pipecat itself, the open source framework described above. The second is Daily's commercial realtime network: a global network of 75 points of presence built for the latency and reliability demands of live AI, according to Deepgram's announcement (Deepgram's announcement). That network is the part that is hardest to reproduce. Writing packet routing software was Daily's original business. So was deploying specialized clusters in dozens of locations, and writing client libraries for web, desktop, and mobile apps. Kramer founded that business with Nina. It is what lets a voice agent answer quickly from wherever the user happens to be.
The third part is expertise. Deepgram says its speech models and realtime infrastructure already power conversational voice AI in production for more than 5,000 organizations, and that more than 650,000 developers have used Deepgram to process more than 200,000 years of audio (Deepgram's announcement). Those are Deepgram's own figures, and they describe scale, not quality. But they explain the logic of the deal from Deepgram's side: it already sells the speech models, and Daily supplies the network and the orchestration layer that turn those models into a working product. Deepgram CEO Scott Stephenson put the same point in plainer language, saying the Daily team deeply understands what it takes to keep a live conversation working reliably.
The two companies also have a long shared history, which makes the deal less surprising than it looks. Daily and Deepgram went through Y Combinator together in 2016, and Deepgram has powered transcription inside Daily's audio and video APIs since 2019, according to Kramer's post (Daily's announcement). An acquisition between two companies that have already worked together for years is a different bet from an acquisition of a stranger: the integration risk is lower, and the people involved already know each other's systems.
Why a Foundation, and Why It Matters
The Pipecat Foundation is an independent foundation that will govern Pipecat. Its steering committee is meant to represent the whole realtime AI ecosystem.
The Pipecat Foundation is the most consequential part of the announcement, and also the least detailed. Kramer writes that the foundation will be managed by a steering committee representing the entire realtime AI ecosystem, including companies that compete with Deepgram. He also writes that Deepgram is adopting Pipecat as the core of its new voice AI products while the framework itself remains vendor neutral (Daily's announcement). Deepgram's announcement makes the same commitment: it will invest in Pipecat's development and community while keeping all of Pipecat open source and vendor neutral, letting developers choose and combine models, tools, transports, and APIs from different providers (Deepgram's announcement).
That structure is designed to solve a trust problem. If one speech company owns the most widely used voice agent framework outright, developers using rival speech models have a reason to worry: will the framework keep working just as well with their models, or will it slowly favor its owner's? A foundation with a committee that includes competitors does not eliminate that worry, but it gives it an address. Developers can watch who sits on the committee, what the foundation's charter says, and whether decisions about the framework are made in public. Until those details appear, the foundation is a promise, not yet a track record, and it should be reported that way.
There is a useful comparison here, and it is not a legal one. Open source projects that survive an acquisition usually do so because their governance survives it: the code stays licensed, the community keeps a real say, and no single company can quietly change the rules. Projects that fade after an acquisition usually fade for the opposite reasons. Pipecat's test will be visible and simple to apply. In six months, can a developer still build a production voice agent on Pipecat using a competitor's speech model, with no Deepgram product in the chain, and get the same support from the project? If the answer stays yes, the foundation is working. If it drifts, the announcement's language will not save it.
The Real-Time Context Standard: What Gets Lost Today
The Real-Time Context Standard (RCS) is an open protocol co-developed by Deepgram and the Pipecat team. It carries audio context between models.
Alongside the acquisition, Deepgram introduced the Real-Time Context Standard, or RCS, an open protocol co-developed with the Pipecat team. The problem it addresses is easy to miss, because it is invisible in a transcript. In a typical voice agent, one model turns speech into text, a language model reads that text, and another model turns the answer back into speech. The words survive that journey. The audio context around them often does not: the speaker's tone, who is speaking, whether a word was unclear, and what is happening in the background (Deepgram's announcement). An agent that only sees words cannot tell a whisper from a shout, or a confident answer from a guess.
RCS defines three kinds of metadata to carry that context along. The first is context from the audio itself, such as a whisper or an unclear word. The second is instructions for how a language model should use those signals. The third is 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 (Deepgram's announcement). In practice, that means an agent could ask for clarification when it did not hear clearly, or answer quietly when the user whispers, instead of treating every turn as identical text.
The demo Deepgram points developers to makes the idea concrete. It is a customer service agent for an outdoor gear shop called Northwind Outfitters. It chains 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. Deployment is on Pipecat Cloud (Unite.AI's coverage). That chain is deliberately mixed vendor: it is the vendor neutrality claim, demonstrated rather than just asserted. RCS is open and extensible, and Deepgram invites developers, model providers, and voice application teams to test it, share feedback, and contribute to the specification (Deepgram's announcement). It is early, and a protocol only becomes a standard when more than one company builds on it, so the right question about RCS is not whether the demo works. It is whether rival model providers adopt it.
What Changes, and What Does Not
The Deepgram Voice Agent Platform is Deepgram's forthcoming integrated path to production for voice agents. It brings Deepgram's models together with Pipecat and Daily's network.
For Daily's existing customers, Kramer writes that nothing will change: Deepgram will sell, support, and develop all of Daily's current products and services, including WebRTC, telephony, and cross-platform libraries (Daily's announcement). For Pipecat developers, the license does not change, the repository stays public, and the framework stays free. For Deepgram, the change is strategic. Pipecat becomes the core of its new voice AI products and services. The forthcoming Deepgram Voice Agent Platform will bring Deepgram's speech to text, text to speech, and speech to speech capabilities together with Pipecat's orchestration and Daily's communications infrastructure (Deepgram's announcement).
Three things remain genuinely unknown, and developers should watch all three. First, the foundation's committee membership and charter have not been published. Second, the purchase price was not disclosed, so the size of Deepgram's bet is not public. Third, the Voice Agent Platform itself is still forthcoming. Deepgram's chief product officer Ed Anuff was scheduled to preview it later on October 29. The preview was to include two new models built for live conversation, expanded language support, and a Global API designed to improve connection reliability and reduce latency. Full technical details will follow on Deepgram's blog (Deepgram's announcement). None of those gaps is a reason to dismiss the announcement. All three are reasons to treat its promises as claims to verify, not facts already proven.
What Independent Coverage Says
Independent coverage on the day of the announcement focused on the same two points: the open source commitment, and the part of the stack Deepgram gains. TechCrunch reported the acquisition on October 29 as a move to strengthen Deepgram's voice AI infrastructure, noting that Pipecat will remain open source while Deepgram gains Daily's realtime network. Unite.AI reported the same day that Pipecat will remain fully open source under its BSD V2 license, that the Daily team will join Deepgram, and that the companies are establishing the independent Pipecat Foundation. Neither outlet treated the foundation as already proven: the reporting describes a commitment and a plan, and the details that would let developers judge it are still to come.
That caution is the right frame for this story. Acquisitions in open source are judged by what happens after the press release, not by the press release. The facts that are already verifiable are strong: the license is unchanged, the repository is public, the framework's scale is documented, and the foundation commitment is explicit and public. The facts that would prove the commitment durable, a published charter, a named and independent committee, and a year of neutral roadmapping, are not yet available. A developer deciding today whether to start a new project on Pipecat is not betting on a promise alone, though: the code they would build on is already open, already licensed, and already forkable, whatever happens next.
Who Should Care, and Who Can Wait
If you build voice agents, this deal matters now, because it changes who maintains the framework you build on and what platform it will be bundled into. Watch the foundation's charter and committee, try RCS if your agents lose context between models, and keep your model choices swappable, which Pipecat's design already encourages. If you buy Daily's hosted WebRTC or telephony services, the practical questions are commercial, not technical: confirm contracts, pricing, and data processing terms as the deal completes, since Deepgram is now your supplier. If you only follow voice AI from a distance, the story to watch is simpler: whether an open source orchestration layer can stay neutral inside a company that also sells the models it orchestrates. That question will be answered by governance documents and release notes, not by launch day statements.
There is also a broader pattern worth noticing. Voice agents are moving from demos to production, and production means the whole system has to work: models, network, orchestration, and context, together. Companies that own only one layer are buying the others. Deepgram's move follows that logic exactly. Whether it helps developers depends on one thing the announcement gets right in principle: the orchestration layer, the part developers build on directly, must stay open and neutral even as the company around it consolidates. Our Scientific Agent Skills article covers a related idea from the other side, skills that change what an agent can do without changing the model underneath. Pipecat is the realtime equivalent: the layer around the model, deciding what the model can hear, say, and do next.
Frequently Asked Questions
Will Pipecat stay open source after Deepgram acquires Daily? Yes, according to both announcements. Pipecat will remain fully open source and free under its existing BSD 2-Clause license, and Deepgram says it will stay vendor neutral. The companies are also creating an independent Pipecat Foundation to govern it.
What is the Pipecat Foundation? It is a new independent foundation that will govern Pipecat, managed by a steering committee representing the entire realtime AI ecosystem, including companies that compete with Deepgram. Its committee members and charter had not been published at the time of writing, so developers should watch for those details.
Does the acquisition change Daily's existing products? According to Daily co-founder Kwindla Hultman Kramer, 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. Those include WebRTC, telephony, and cross-platform libraries.
What is the Real-Time Context Standard? It is an open protocol co-developed by Deepgram and the Pipecat team that carries audio context between models, instead of passing only transcribed words. It defines three kinds of metadata: context from the audio, instructions for how a language model should use those signals, and guidance for how the response should sound.
How much did Deepgram pay for Daily? The companies did not disclose a purchase price. Coverage on October 29, including TokenPost's report, confirms the price was not shared, so any figure circulating without a source should be treated as unconfirmed.
Related reading: our Harness article covers turning one prompt into a full AI dev team, and our Agent Scripts article covers sharing one set of AI coding skills across Claude Code and Codex.
How this article was checked: we analyzed Deepgram's announcement and Daily's announcement, then compared them with TechCrunch and Unite.AI coverage. This article passed editorial review before publication.
The Bottom Line
Deepgram's acquisition of Daily is a bet that voice AI has grown up: the demo era is over, and the winners will own the whole production stack, from models to network to orchestration. The encouraging part is that the companies know the orchestration layer only works if developers trust it. They have structured the deal around that trust. The license is unchanged. The foundation is meant to outlive any one vendor's strategy. The honest part is that trust is not yet earned. It will be earned, or lost, in the foundation's charter, its committee, and the next year of releases. Start with the Pipecat repository, read Deepgram's announcement and Daily's announcement, or read the Indonesian version of this article. If you build on Pipecat today, nothing in the license forces you to change. That, more than any launch day quote, is the fact that matters.