The judgment layerfor AI agents.

Teams are deploying agents faster than they can govern them. The work gets done, the spend gets spent, and nobody can show how the decisions were made.

Liminal is the layer that makes delegated work legible: who decided, on what basis, corrected by whom, and what it cost. Local-first by architecture, not policy. Your record lives in your vault, not our cloud.

Liminal shipped its first product, a structured-reflection tool, in 2024. In March 2026 it was rebuilt around a harder version of the same conviction: the decisions people make with AI deserve infrastructure. The architecture of that rebuild is patent-pending.