What Delentia Labs has already proven — in working code, passing tests, and documented benchmarks.
One person who does not write code, directing AI coding assistants. The result so far: an open-source agent runtime whose public repository CI runs 7,820 automated tests. What it shows is narrow: one person can build and test a governed agent runtime this way. It does not show that the runtime works well with a capable model or for other people; that has not been measured yet.
The public repository CI covers 7,820 automated tests across unit, integration, and end-to-end layers. Backend-validated coverage sits at 66.7% with a public 100% target. The test count is not a vanity metric — it is the organization's commitment to evidence culture, visible and trackable from day one.
No GAIA or other leaderboard score is claimed. An earlier version of this page quoted a projected range, which had no measurement behind it and has been removed. A result will be published only after it has been measured and submitted.
Delentia began in Bangkok. The aim is a governance layer whose rules the owner writes (a data-residency policy, who signs what), so that the same runtime can be used under different national rules. It does not know any country's law and certifies no compliance; that remains the owner's and their counsel's job.
Every claim on this page is backed by documented benchmarks, test suites, and public specifications.