Today we're opening the waitlist for the Cobble beta.
After months of building routing software, metering systems, and a distributed inference network assembled from reclaimed hardware, we're ready to start putting Cobble into the hands of real users. If you want sustainable, OpenAI-compatible inference that doesn't cost the earth, this is your invitation to get in line.
How to Join
Joining takes less than a minute.
- Head to the Get Started page.
- Enter your email and a few quick details so we can tailor your onboarding.
- Submit the form — that's it. You're on the list.
The beta is currently open to U.S.-based subscribers. If you're outside the United States, you can still leave your email, and we'll notify you the moment Cobble becomes available in your country.
What Happens After You Sign Up
We are approving new subscribers in small, deliberate batches rather than all at once.
When your turn arrives, you'll receive an invite email containing a secure link to finish setting up your account, choose a plan, and generate your first API key. From there, you can point any OpenAI-compatible tool at Cobble by changing the base URL and key — and you'll be running on recycled silicon and renewable energy.
Keep an eye on your inbox. Invites go out in waves, and each wave opens up more capacity.
Why We're Onboarding in Batches
It would be easy to open the floodgates. We're choosing not to.
Cobble runs on reclaimed GPUs and refurbished server hardware, deployed thoughtfully across regions with access to renewable energy. Growing in measured steps lets us protect performance, reliability, and support quality for every subscriber — instead of chasing raw signups at the expense of the experience.
Small batches also let us learn quickly. Each wave of users helps us tune routing, refine limits, and harden the platform before the next group arrives. When you join, you're not just a customer — you're helping shape a more sustainable kind of AI infrastructure.
What You Get in the Beta
- Access to a growing catalog of open-weight models — generative, OCR, and embeddings.
- An OpenAI-compatible API that works with the tools you already use — see the docs.
- Transparent, predictable pricing with flat-rate plans and pay-as-you-go wallet top-ups.
- Inference served from reclaimed hardware, so scaling up never means a bigger footprint.
Reserve Your Spot
The most powerful systems of the future can be built from the materials the world almost threw away — and you can be part of proving it.
Join the waitlist today. We'll be in touch soon.