Today weโre announcing LiveKitโs $22.5M Series A to build the transport layer for AI.
This wasnโt an easy fundraise. Late last year, we pitched investors that realtime voice and video would become THE way we interact with computers. A few didnโt agree; most said it was at least
Today weโre launching LiveKit Cloud, the result of a year of talking with devs building live video and audio products with our open-source WebRTC stack. Hereโs what we learned:
So, if we're building the metaverse, it needs a communications stack. LiveKit, our open source WebRTC infrastructure, brings audio and video to the next generation of apps. I'm pumped about what we're building next:
For founders asking:
OSS is not a strategy, itโs an ethos.
We started LiveKit not thinking it would even be a company.
@davidzh
and I just wanted to pay it forward for the shoulders weโve stood on the last 20 years of our careers.
We got lucky that companies wanted to pay for
Two questions Iโve gotten lately:
1. how much AI will be done locally vs. in the cloud?
2. do you still need
@livekit
if you can run a multimodal model on device?
Short answers:
1. 80/20
2. Yes
My thoughts:
- In humans the triune brain does a lot. But I still reach for my
The Character team is legit.
They train all their own models from scratch for asr, stt and tts. Was super fun for the
@livekit
team to work with them on this.
๐ฉโ๐ค๐งโโ๏ธ
AI Chat just got real.
Introducing Character Calls, the latest addition to our suite of Voice features. Talk with AI Characters, just like you would talk on the phone. Whether it's polishing your language skills, acing interviews, adding zest to RPGs, or crafting epic
A global network for live audio/video. A novel pricing model that unlocks entire classes of new apps. And best-in-class media analytics.
Are you getting it?
Today everyone can take a live audio/video app from prototype to production using LiveKit Cloud:
Me: Yasssss!
Mom: Let me guess, some stupid software thing you figured out?
Me: Mom, it'd take you 100 years to figure this out.
Mom: Nope, I'd go on GITHUB and figure it out in a couple minutes.
14y ago, new Twitter employees could claim an "inactive" (6mo since last tweet lol) @-username. I was the only one in my cohort that went straight for the single-letter options.
@x
,
@u
,
@i
,
@m
,
@y
and a few others were available. I tried
@u
for a bit, but notif spam was insane!
To pull this off, we've assembled a seriously ridiculous team, with deep experience in audio, video, distributed systems, networking, cloud gaming and netcode. We also raised $7m from amazing folks that share our vision. LFGGG! ๐
In this episode of the Autopilot podcast, host
@WSummerlinAI
interviews Russell
@dsa
, the founder of
@livekit
, a company building critical infrastructure for real-time audio and video communication, used by OpenAI to voice ChatGPT.
Russ shares his journey from growing up
โSmallโ is an understatement.
@gather_town
was our first large scale customer and stuck with us through the early days of bugs and outages.
We wouldnโt be here without your support
@k_m_y_l
and
@flippnflops
.
Congrats on the raise guys ๐
Easily one of the most stacked teams Iโve ever interacted with. And with an expansive vision.
Honored to be some small part of the story.
So many takes about how Vision Proโs software is isolating or underwhelming. Did we forget Apple wrote the book on consumer electronics GTM? Itโs not just the hardware either; their software roadmap is 5-10 years ahead.
Early adopters first, with single-player apps and bridges
Devs love working with LiveKit, but running a real-time video network at scale in prod is hard. Most teams would rather focus on finding product-market fit.
So we built a global SFU mesh, where users connect to edge nodes, and most data travels over backbone connections.
The industry-standard participant-minute pricing model for live audio/video is cost-prohibitive for certain use cases, making entire classes of apps economically unviable
So we designed a cost-plus pricing model where you only pay for the bandwidth and compute your app uses
Mom: Where did โBye Feliciaโ come from?
Me: I donโt know, but I work with a Felicia, so sometimes we say that.
Dad: Oh, is he Iraqi?
Me: She is not. Wut?!
Dad: Thereโs a city in Iraq named Felicia.
Me: Thatโs FALLUJAH!
๐คฆ๐ฝโโ๏ธ
Mom: Iโm checking out Hacker News right now. How do I read comments?
Me: Click on the comments link. Youโre probably not gonna understand them though.
Mom: Weโll see about that.
I shouldnโt doubt someone who once told me to, โgo to GitHub and find some code to fix your bug.โ
@tanncap
@SquareCash
@jack
Way back in the day,
@jack
's real account was used in staging. We made him follow a few accounts, not sure if he noticed. ๐
"With Cerebras, you can get inference down to 50-100 milliseconds. It's incredible how human-like conversations you have with an AI model can be with fast inference." -
@dsa
CEO at
@livekit