Deepgram is acquiring Daily, the company that builds and maintains Pipecat, one of the most popular open source frameworks for voice AI. TechCrunch AI reports that the deal was announced Thursday. Neither company disclosed the price. Both say Pipecat will stay open source.
This is another consolidation move in voice AI, a sector where money is coming in fast on both the model side and the infrastructure side.
⚡ Key Takeaways
- Deepgram moves up the stack. It’s best known for speech models. Now it owns the orchestration layer that connects those models to everything else.
- Pipecat stays open and vendor-neutral. Deepgram CEO Scott Stephenson committed to keeping it open source “forever,” according to Daily co-founder Kwindla Kramer.
- There’s a new open protocol. Deepgram is also launching the Real-Time Context Standard (RCS), which it built together with the Pipecat team.
- More consolidation is coming. Voice AI infrastructure is crowded, and the bigger players are buying their way into stronger positions.
🎙️ What Pipecat Actually Does
A voice agent isn’t one model. It’s a chain. Audio comes in and speech-to-text (STT) turns it into words. A large language model works out a reply, and text-to-speech (TTS) speaks it back. Every handoff adds delay, and in a live conversation even small delays make the agent feel off.
Pipecat handles that chain. It manages the voice and video streams and coordinates the STT, LLM, and TTS pieces, all while working to keep latency low. Daily also runs hosted infrastructure for voice and video calls.
Kramer told TechCrunch that the team started experimenting with AI-to-human communication over voice and text around the time GPT-4 came out, and Pipecat grew out of that work. “We never thought about Pipecat as a voice toolkit. We thought about it as the toolkit we needed to build this new kind of AI-native software,” he said.
🤝 Why Deepgram, and Why Open Source
Kramer said several companies were interested in buying Daily. Deepgram won out because of Stephenson’s commitment to open source and because the two engineering cultures fit. There’s a bit of history here too: both companies came out of the same Y Combinator batch in 2016.
What stands out is how openly both sides talk about the business logic. Kramer put it this way:
“If you sit in the middle of an ecosystem that’s growing very fast, and have many parts of that ecosystem that are genuinely the best from your company, then everything being as open as possible is like a commercial advantage.”
Stephenson went further and said he’s fine with developers using competitors’ models. “It’s great if you’re using Deepgram models, but I want you to be using the best models for your use case,” he said. If Deepgram falls behind in an area, it’ll see that and work to catch up.
That’s a confident bet. If Deepgram owns the framework everyone builds on, it gets a front-row view of which models developers actually choose and why.
🧩 The New Protocol: RCS
The Real-Time Context Standard is an open protocol for voice agents. It defines metadata that travels with the conversation, including:
- Audio context, such as whether the speaker is whispering
- Instructions for the LLM on how to use those signals
- Guidance on delivery, meaning how the response should sound
This goes after a real weakness. Most voice pipelines throw away everything except the words. Tone, volume, and hesitation get lost at the STT step. RCS tries to carry those signals through the whole chain so the agent can respond to how something was said, not just what was said.
💰 The Money Behind It
- Daily has raised over $62 million from investors including Tiger Global, Highfield Capital, Nat Friedman, Lachy Groom, and Kevin Weil, according to PitchBook data.
- Deepgram became a unicorn earlier this year after a $130 million Series C.
- Model makers like ElevenLabs, Gradium, and Fish Audio have raised millions.
- Infrastructure players LiveKit and Vapi both closed multimillion-dollar rounds this year.
🔭 What This Means for Builders
If you’re building on Pipecat today, nothing breaks. The framework stays open, and you can still plug in whatever models you like. Keep an eye on two things over the next few months: whether Deepgram integrations start getting noticeably smoother than the alternatives, and whether RCS gets picked up outside Deepgram’s orbit.
Stephenson also hinted that Pipecat gives Deepgram a way past pure voice, since the framework already handles real-time communication in several formats. The bigger prize here isn’t speech recognition. It’s becoming the default plumbing for real-time AI agents. Full details are in TechCrunch AI’s original report.