Selah
for investorsthesis · stage-honest

Decide before any agent acts.

We build the layer that decides what AI agents may do before they act. Agents propose, our engine decides, connectors execute, and the agent never holds a key.

The thesis

Every agent that touches money, data, or customers will need a gate in front of it. Observability tells you what happened; regulators, insurers, and operators increasingly require control over what may happen. We build that control as infrastructure: deterministic, fail-closed, under ten milliseconds, with compliance packaged as enableable packs across jurisdictions. The product, reduced to one sentence: we make AI agents trustworthy by structure (no keys, a gate, a record), which is what turns hallucination from a liability into a non-event.

Where we are

Our engine is live, our three checks are integrated through three endpoints, our shadow mode runs against real traffic, and our compliance library is being authored jurisdiction by jurisdiction across Latin America first, by design: multi-jurisdiction production history is the moat when the category globalizes. Stage-honest: we are pre-seed, building in the open with design partners, and we publish no metrics we do not have.

The category

Industry analysts began naming agent governance and management as a category in 2026. We sit at its prevention end, the only end that stops a loss instead of narrating it.

Request the memo.