Brev.dev is a tool to easily get a GPU for fine-tuning or training. We want to be the simplest developer experience to build any AI/ML application, but ensuring developers are free from vendor lock-in. The design of our product from the beginning ensures developers own the entire stack, and Brev.dev as a tool makes that possible. If it’s not your compute, it’s not your platform!
There’s a squeeze on GPUs, but we’ve found a way to reliably get access to single node GPUs. This means 1-8 GPUs, which is more than enough for fine-tuning something like Llama2. This unlocks a ton of use cases. In the time frame of the squeeze, which we expect to last 18-24 months, we can build out an entire software pipeline. There’s a unique opportunity to accelerate years of growth because of the squeeze and we’re taking advantage of it.
We feel there’s a moral imperative for AI to stay open-source and to ensure that teams own the full stack of their application. Wrappers of GPT-4 and other closed-source APIs are a really effective way to build valuable products, but it’s dangerous to accept such strong platform risk.
Additionally, AI is a strong centralizing force. It’s inevitable that AI becomes more personal and has access to all of your personal data. It’s scary to think that one company might own large swaths of everyone’s data. We want to empower individuals to create their own AI, but without forcing them into high abstractions.
We’re doing everything we can to build an impeccable product. Usage is growing. Demand is growing. And every step we take shows us 3 more we can take.
We're a team of second-time YCombinator founders, infrastructure and ML experts, building on our experiences at Microsoft, Workday, Meta, Stanford, and previous startups.
We are fortunate to have investors that include Brickyard Ventures, 468 Capital, Chris Klaus, Stephanie Sher, Swyx, and HustleFund as well as founders and operators from Datadog, Github, ContainIQ, Carbon Health and Bellhop.