The future runs on open weights.
An open-weight model is one you can download and run yourself. That is what makes private inference possible. You cannot attest, audit, or air gap a model you are not allowed to hold.
Nobody can change it under you
No vendor can deprecate, reprice, or quietly swap the model your business depends on.
Fitted to your hardware and your work
Tune, quantize, and serve the model for the GPUs you have and the job you need done.
The gap keeps narrowing
Open models trail the closed frontier by less each year. For many business workloads, they are already enough.
Open source, all the way down.
This is a conviction, not a cost saving. The internet was built on software anyone could read, fix, and share. We build on open source first, and reach for something closed only when nothing open can do the job.
You can read every line
Security claims you can check beat claims you take on faith. Open code is the only kind you can audit.
You are never locked in
No license can be pulled or repriced at renewal. What we build keeps running whether or not you keep us.
Giving back
We have presented at TalosCon 2023, appeared on the Kubernetes for Humans podcast, and published open source work of our own.
What we choose: Talos Linux
We work with whatever Kubernetes you already run. Ask what we would choose, and the answer is Talos Linux from Sidero Labs.
We adopted it more than five years ago, when an operating system with no shell and no SSH, managed entirely by API, still sounded radical. It turned out to be where the industry was heading. Few projects have stayed so consistently ahead of their time.