Daily.co

@blockdevninja hey Lionel - Daily Bots isn't an end-user copilot tool like ChatGPT or Jasper. We're focused on allowing developers to build, host, and scale new conversational AI experiences on Daily's cloud and leverage an open-source tech stack.

@dance17219 Thank you! It's been really interesting swapping out the models!! Curious to hear from the community what models they're using.

@ema_elisi Thank you! Our team is really excited to ship this and hear from the community about this! 🍻

@mert_gerdan Standardization is definitely a big part of the motivation for Daily Bots overall. - We want developers to be able to write code that they can use with a wide variety of real-time inference services. (Not just Daily.) - We think that standardization helps an ecosystem grow faster. - The more inference providers have really good real-time offerings, the more people will understand the nuances of low-level networking, how to optimize for low latency, etc. And that's the area we've always focused on building great tech for!

@pj_mukh Implementing scalability elegantly was one of our main goals for Daily Bots. We have a lot of experience running infrastructure, so I'm confident we can scale efficiently -- both so individual customers see fast bot joins and so we can charge low prices that enable a broad range of use cases. Having said that, forking Python processes is actually part of how we achieve fast join times. Avoiding the Python interpreter startup time is helpful! But we also have capacity routing via queues, autoscaling that leverages the cluster scaling we do for WebRTC at Daily, etc.

@robertthomas2 +1, it'll be really fun to see what makers do with this! Thank you, and it's nice to be back on PH :)

@patriciaharris Daily Bots can deliver voice-to-voice latency as low as 500ms, depending on the models you choose. This is substantially faster than most other voice apps. Typical latencies are 1.2 seconds or more. There are trade-offs: using larger and more capable LLMs tends to be slower than using smaller, faster models. However, we've partnered with Together to make it easy to use the Llama 3.1 models with Daily Bots and together has very fast "time to first token" metrics even with the largest (405B) Llama 3.1 model. Daily Bots is brand new. But we've worked closely with many partners who have built on Daily and Pipecat to create the pieces that Daily Bot brings together into a single platform. You may have seen Tavus' launch last week. Tavus was #1 product of the day last Thursday. If you're interested in the open standards that Daily Bots is built on, check out Pipecat and RTVI: - https://github.com/pipecat-ai/pi... - https://github.com/rtvi-ai