Rules are deterministic. AI proposes, policy enforces.
Every rule in DataInbox compiles to a deterministic policy. AI helps you write and suggest rules, but never decides enforcement. Routing, validation, and approval are explicit, versioned, and fully auditable.
Routing rules
Decide which inbox, agent, or system receives the event.
IF event.type = order.created AND event.amount > 1000 THEN route to high-value-orders inbox
Validation rules
Enforce schema, business invariants, and data quality at ingestion.
IF event.currency NOT IN [EUR, USD, GBP] THEN reject and notify source owner
Approval rules
Require a human or expert to confirm before an AI proposal becomes an action.
IF agent.action = refund AND event.amount > 500 THEN require approval by finance
Authoring a rule
- 1. Describe. Type the rule in plain English in the AI Rule Builder, or write YAML directly.
- 2. Preview. See the rule evaluated against the last 1,000 matching events before it is saved.
- 3. Approve. Reviewer approves the diff. The rule is versioned and signed.
- 4. Roll out. Promote to all traffic or run as a governed A/B test against the current rule.
- 5. Audit. Every evaluation is logged with input, decision, and rule version.
Rules FAQ
How are rules defined in DataInbox?
Rules are authored in a plain-English builder that compiles to a deterministic policy. Every rule has a trigger (event pattern), conditions (field tests), and an action (route, validate, approve, transform). Engineers can also write rules in YAML or via the API.
Are rules deterministic or AI-driven?
Rule execution is fully deterministic. AI is used only to suggest new rules and to draft proposals. Every enforcement decision is made by the deterministic policy engine, never by a language model.
Can non-engineers write rules?
Yes. The AI Rule Builder accepts plain-English descriptions and turns them into structured rules with previewed effects. Business users author and edit, engineers review and approve.
Are rules versioned?
Yes. Every rule change is versioned, attributed to a user, diffed, and replayable against historical events. Rollback is one click. Rule changes also produce an audit log entry.
Can I A/B test a rule before rollout?
Yes. Governed A/B testing lets you run a candidate rule on a percentage of live traffic, compare outcomes against the control rule, and promote with full audit trail.
What governance controls are needed for agentic AI?
Three controls are essential: deterministic policy enforcement that AI cannot bypass, scoped data access per agent and per schema, and immutable audit logging of every event, proposal, approval, and override. DataInbox provides all three by default.