Pinned Tweet
When @jess & I were launching @Adverbvc, we were eager for a day we could proudly display the humans we’d backed on our website.
This month, we’re celebrating our two-year anniversary as a firm. We've backed 22 incredible companies, added 2 new team members (@LizaGurtin & @gary_pesola) & have a new website to boot. Check out the companies we’ve backed, and subscribe to our Substack for updates.
adverb.vc
🚀
This month, we’re celebrating our two-year anniversary as a firm. We've backed 22 incredible companies, added 2 new team members (@LizaGurtin & @gary_pesola) & have a new website to boot. Check out the companies we’ve backed, and subscribe to our Substack for updates.
adverb.vc
🚀
9
3
56
28,115
April Underwood retweeted
Today we launched Kinference, a new open source project I've been working on for building scalable inference platforms on top of @vllm_project. Self-hosted inference is such a fascinating and crazy space and I did not really expect to ship a project here, but this is probably the most excited I've felt about anything since the early days of @Linkerd!
Kinference came about when we saw how hard it was (for our customers and ourselves) to run your own inference. Sure you can serve a model, but... what then? What if you have a bunch of models? And you want to change them without interfering with your clients? What if you have multiple users with different requirements and you need to juggle them all? Make sure no single user can consume everything? Make behavior under saturation or failure explicit? What if you need to burst to the cloud? And on and on...
Once we took more than 3 steps down this path it was clear that serving inference was a *platform-shaped problem*. Yes shockingly no one had built a platform-shaped solution. Which is crazy because platforms are central to how to build scalable reliable software in the modern world!
So we ended up building Kinference. We design it to be a platform layer for inference. It runs on top of vLLM and provides all the features necessary for treating it as a platform problem, including token rate limiting, tenant usage and cost tracking, intelligent load balancing, and inference policies. The policy stuff is especially powerful IMO and I'm pretty happy with how it turned out.
It's an early project, but we've been using it internally at Buoyant to juggle a bunch of AI tasks from a bunch of clients across a bunch of models, and it has been awesome.
So: if you are serving inference to an ever-changing set of clients with varied and competing demands, from a ever-changing set of LLMs running on a heterogeneous, evolving pool of GPUs, and somehow have to keep everyone happy... please take a look and I'd love your feedback!
kinference.ai
Kinference came about when we saw how hard it was (for our customers and ourselves) to run your own inference. Sure you can serve a model, but... what then? What if you have a bunch of models? And you want to change them without interfering with your clients? What if you have multiple users with different requirements and you need to juggle them all? Make sure no single user can consume everything? Make behavior under saturation or failure explicit? What if you need to burst to the cloud? And on and on...
Once we took more than 3 steps down this path it was clear that serving inference was a *platform-shaped problem*. Yes shockingly no one had built a platform-shaped solution. Which is crazy because platforms are central to how to build scalable reliable software in the modern world!
So we ended up building Kinference. We design it to be a platform layer for inference. It runs on top of vLLM and provides all the features necessary for treating it as a platform problem, including token rate limiting, tenant usage and cost tracking, intelligent load balancing, and inference policies. The policy stuff is especially powerful IMO and I'm pretty happy with how it turned out.
It's an early project, but we've been using it internally at Buoyant to juggle a bunch of AI tasks from a bunch of clients across a bunch of models, and it has been awesome.
So: if you are serving inference to an ever-changing set of clients with varied and competing demands, from a ever-changing set of LLMs running on a heterogeneous, evolving pool of GPUs, and somehow have to keep everyone happy... please take a look and I'd love your feedback!
kinference.ai
13
17
97
5,309
April Underwood retweeted
New today: Casa Plus, at $99/month in SF, LA, and Austin.
It comes with an hour of handyman time per month that rolls over, weekly package pickups (a fan favorite), Autopilot, and a bunch more to help you care for your home.
We also shipped a lot of app updates. You can now message, call, and video chat your handyman, easily book a follow-up visit, and go through a beautiful flow to add To-Dos.
Oh, and dark mode. I'm a light mode guy myself, but go for it.
getcasa.com/updates/plus-a…
x.com/getcasa/status…
It comes with an hour of handyman time per month that rolls over, weekly package pickups (a fan favorite), Autopilot, and a bunch more to help you care for your home.
We also shipped a lot of app updates. You can now message, call, and video chat your handyman, easily book a follow-up visit, and go through a beautiful flow to add To-Dos.
Oh, and dark mode. I'm a light mode guy myself, but go for it.
getcasa.com/updates/plus-a…
x.com/getcasa/status…
Everyone deserves a joyful homeownership experience.
Introducing Casa Plus, our new home membership packed with everyday home help for just $99/month. It includes:
- An hour of handyman time every month that rolls over
- Weekly package and donation pickups
- Autopilot for your home's recurring maintenance
- Mailbox, so you can read your physical mail from anywhere
- Trade discounts at the home stores you already shop
- and much more...
4
3
36
4,097
👻 Not All Venture Capitalists 👻
Venture capitalists love to talk about pivoting. Good. They're gonna have to.
axios.com/2026/11/04/ven…
1
0
21
1,376
And by the way, I'd love to see some analysis on the gender split between VCs who swung far right, and those who stayed true to their values even when unpopular (and even if quietly). I can tell you the answer, but I'd like to see the data. :)
0
0
5
213
See the thing is, y'all thought you knew what Texans were about and you were dead wrong. And damn it feels good.
0
2
13
1,052
How your email / slack / text finds me: youtu.be/BodXwAYeTfM?is…
0
0
2
899
As if the blue tsunami wasn’t enough, the texts I’m getting from people who know how much this night means to me have put me over the top
2
0
93
3,968
