policy as code
Policy authoring
VisIQ policy is one rule shape over three operations: retrieval, action, and delegation. You draft a rule in plain terms, simulate it against recent recorded traffic, and enforce it once the simulation shows it leaves legitimate work alone. Every decision receipt is bound to the policy version that produced it.
One rule shape over every operation
VisIQ governs agents with one rule engine acting on events. Each event carries an operations tag drawn from retrieval, action, and delegation, and a rule declares which operations it applies to. A hybrid tool that reads and then writes is one event tagged with both, evaluated once.
Every rule emits an outcome from the same vocabulary: permit, deny, approval required, redact, escalate, or mask. Rules evaluate in priority order and the first decisive outcome wins. Where nothing matches, your configured no-coverage default resolves the event, and that default is itself a deliberate choice: permit, require approval, or deny.
Written in your organization's terms
The default rule set crosses two agent attributes into a need-to-know matrix: a business function such as finance or engineering, inferred from the agent's observed traffic and pinnable by an operator, and a trust tier that is always a human decision. One matrix governs the knowledge, action, and approval boundaries at once, with no per-tool wiring.
An accounting agent sees a full SSN; an engineering code-review agent sees it masked. That intent takes two attributes to express, and when the organization changes, a function pin or a retier reaches every running agent on its next rule-bundle refresh.
Test a rule against real traffic before it can break anything
New agents run in monitor mode first, and every tool call plus the decision it would have received lands in the audit trail. Within days you hold a factual record of what your agents actually do: which tools, which arguments, at what volume.
Every new or edited control is simulated against recent real traffic before it takes effect, so a rule that would inhibit legitimate work is caught before it ships rather than after it breaks the agent. Actions that no rule covers land on a review queue, each paired with a draft control that would cover them.
Versioned, reviewed, provable
Review is built into the authoring path. AI-drafted controls are created disabled and a human enables them, so no agent is ever governed by a rule nobody approved.
Each enforcement receipt names the posture, policy version, and enforcement basis behind the decision it records. When an auditor asks why an action was permitted last March, the answer is the rule that was live at that moment, in that version, and the signed record that cites it.