<iXi> inixiative / demos

Interactive demo · work in progress

The engine for your AI SaaS.

Your application’s conditions grow: nested exceptions, customer-specific access, questions written by agents. Express those conditions as JSON and reuse the engine to evaluate data, compile supported database queries, and enforce each agent’s boundaries. You build the application; the engine handles the changing rules.

Open the workbench  ↗

Public sandbox · sample data · no sign-in

Learn the engine, step by step

Start with Video 0 · narrated · English captions

Rules engines: start with one condition

Read the lesson transcript

Start with the concept lessons, then watch The engine in action: an actual workbench recording with lenses, related data, two execution paths, and changing boundaries. Synthetic-fact inference in the concept lesson uses a prepared AI example.

  1. 00 / RULES ENGINES

    Start with a condition

    See records pass or fail one test. Learn what a predicate is, how groups combine conditions, and how the engine evaluates data or translates a supported rule into a database query.

  2. 01 / LENSES

    Define the data boundary

    A lens describes the fields and records a query can use. Narrow it, bind an identity, and follow a nested predicate into a database query. Try to cross the boundary and see validation reject it.

  3. 02 / PERMISSIONS

    Decide who receives it

    A permission policy decides who can receive a lens. Combine attributes, roles, and relationships in JSON. Change access upstream and watch the available lenses and query results change.

  4. 03 / SYNTHETIC FACTS

    Make an interpretation queryable

    Infer something the original fields don’t say. Represent it as a typed fact with evidence, connect it through a bridge, and write one predicate across source records and derived interpretations.

A stale rule no longer fits its lens

A rule is built against a lens. If upstream narrowing removes a field or relation the rule uses, that rule is no longer valid: it is stale. The validator shows the broken reference, and the predicate must be revised to fit the current lens. Fewer matching records, old results, and expired facts are separate concerns.

Try it yourself

Open the workbench and choose Run the demonstrations. Four guided flows execute against the same sample records: rules, lenses, permissions, and synthetic facts. Edit the JSON, run each step, inspect the real result, then change a boundary and see what must be repaired. The three columns reflect the configuration you just changed.

Shipment records are sample data; the engine applies to your own schema and domain. The videos introduce the concepts; the working demonstrations let you operate them. Ask “Which lenses can I use?” or “What values does status accept?” for current metadata without a model key. The public preview uses guided inference examples with real engine validation and execution. Open-ended AI questions are not enabled.

For developers: the connection as JSON 
{
  "endpoints": [
    { "fieldMap": "demo", "model": "Shipment", "on": "id" },
    { "fieldMap": "facts", "model": "DerivedFact", "on": "shipmentId" }
  ],
  "cardinality": "oneToMany"
}