About
Why Norvia Labs exists
We build and publish focused open-source experiments so the agent stack gets better in public — not only inside closed notebooks.
The thesis
The agentic era is not short on demos. It is short on boring infrastructure that survives contact with production: sandboxes that fail closed, traces you can actually read, cost maps that match invoices, fixtures that keep CI honest.
Norvia Labs is a lab for those pieces. We treat each project as an experiment with a clear question, a small surface area, and a public outcome — ship, learn, or archive.
What “experiment” means here
An experiment is not half-finished product marketing. It is:
- A single sharp idea with a README that states the bet
- Runnable code under MIT
- Enough metrics or fixtures that someone else can reproduce the claim
- Permission to be wrong in public
How we work
Public by default. Design notes and dead ends stay visible when they help the next person.
One idea per repo. Composition beats monoliths. If two concerns fight for the same README, they should be two experiments.
Measure first. Latency, cost, failure modes — if we recommend a tool, we want numbers next to the claim.
Families
Work is grouped into Foundations, Runtimes, Observability, Tooling, and Research. That taxonomy is a map, not a prison — experiments can move families as they mature.
Contact
Open an issue on GitHub or use the contact form.