The semantic layer
writes itself.

Paste a schema. Every column comes back with what it means, how sensitive it is, and the trap an LLM would fall into using it unaided.

Try it on a schema

Your schema

CREATE TABLE building_units (
  unit_id           BIGINT PRIMARY KEY,
  property_id       BIGINT NOT NULL,
  unit_number       VARCHAR(16) NOT NULL,
  floorplan_code    VARCHAR(24),
  sqft              INTEGER,
  market_rent_cents INTEGER NOT NULL,
  status            VARCHAR(24) NOT NULL,
  offline_reason    VARCHAR(64),
  ready_date        DATE
);

CREATE TABLE leases (
  lease_id            BIGINT PRIMARY KEY,
  unit_id             BIGINT REFERENCES building_units(unit_id),
  primary_resident_id BIGINT REFERENCES residents(resident_id),
  start_date          DATE NOT NULL,
  end_date            DATE NOT NULL,
  rent_cents          INTEGER NOT NULL,
  concession_cents    INTEGER DEFAULT 0,
  concession_months   SMALLINT DEFAULT 0,
  renewal_of          BIGINT,
  terminated_at       TIMESTAMPTZ
);

CREATE TABLE residents (
  resident_id     BIGINT PRIMARY KEY,
  full_name       VARCHAR(120) NOT NULL,
  email           VARCHAR(160),
  phone_e164      VARCHAR(20),
  ssn_last4       CHAR(4),
  screening_score SMALLINT,
  created_at      TIMESTAMPTZ NOT NULL
);

CREATE TABLE tour_events (
  tour_id       BIGINT PRIMARY KEY,
  unit_id       BIGINT REFERENCES building_units(unit_id),
  channel       VARCHAR(24),
  agent_user_id BIGINT,
  scheduled_for TIMESTAMPTZ,
  attended      BOOLEAN,
  outcome       VARCHAR(32)
);

Semantic contract

Nothing yet.
Run it and every column comes back documented.

Runs live. Swap in your own tables.

The semantic layerwrites itself.

The semantic layer
writes itself.