Today we're launching Arbr.
Every team building with AI right now is quietly giving up control. Over which models they run. Over their data. Over what happens when a provider changes a price, deprecates a model, or ships an update that breaks production.
Arbr is about taking that back. We're calling it what it is: independence.
Independence from model lock-in. Route every call across any provider. Switch models without rewriting your stack.
Independence from vendor lock-in. Self-hosted, MIT licensed. Runs on your own infra, not a SaaS you rent.
Independence from model churn. New models ship every week. Shadow test and adopt them safely, on your terms, without betting production on a version bump.
Independence from flying blind. See every call. Score quality on live traffic. Turn what you learn into training data you own.
Arbr is the open control plane for AI workloads. Route, govern, observe, evaluate, capture, deploy. All self-hosted, all open source.
The closed stack was never going to give you independence. So we built the open one.
This is year one. Go try it, star it if it resonates, and tell us what breaks.
LINK IN THE FIRST COMMENT.