How it works

Constellation appears only when an action crosses an authority boundary, threshold, timing rule, or irreversibility cliff.

An AI agent about to send an email to a regulator. A team member using Claude to draft a procurement response. A manager approving a financial commitment between meetings. These are moments of consequence — where institutional knowledge needs to be present regardless of who or what is acting. Constellation intervenes here and only here.

Everything else passes through silently.

Before a consequential action is taken, Constellation checks it against the institution’s constraints — authority rules, spending thresholds, timing restrictions, prohibitions — and returns applicable constraints, violations, and required approvals.

For AI agents, this happens automatically via MCP integration. For humans using AI tools, Constellation is present in the workflow — the same constraints apply whether a response was drafted by Claude or written by hand. The institution’s own rules are consulted before anyone acts.

Authority, escalation, and acknowledgment are carried by the institution itself — not improvised by people under pressure. Constraints are defined in advance through the Charter, not invented at the moment someone needs to decide.

When a constraint is violated, the institution’s own structure determines who must approve, what authority is required, and how long they have to respond. The governance is in the architecture, not in someone’s memory.

Every consequential action produces a governance trace: what happened, who authorised it, what constraints were checked, whether any were violated, and what influenced the decision.

This evidence is produced in real-time, not reconstructed later. When an auditor asks “under whose authority did this happen?” the answer already exists.

Most governance tools assume compliance. Constellation assumes disagreement.

Any institutional decision can be formally challenged through the Forum. Challenges require standing and evidence. Adjudicators issue rulings. Rulings carry precedent weight that influences future decisions. This is a common law system inside a product.

AI delegation is not a switch you flip. It is trust earned through evidence.

In Shadow Mode, AI analyses events and recommends actions. The institution decides while Constellation tracks agreement rate per decision category. When accuracy exceeds the threshold with sufficient sample size, delegation becomes available. Delegation can be suspended instantly with full rollback.

You don’t turn on AI. You calibrate trust through evidence.