Core concepts

Same vocabulary in SQL, Slack, and Snowflake views—debug by following the lifecycle.

Policy is evaluated deterministically before execution. Intents, decisions, approvals, and receipts are system-of-record data in your Snowflake account—not an opaque platform log outside your warehouse.

Lifecycle

Request → Policy → Approval → Execution → Receipt

Action — typed work (action_key + params).
Context — optional row(s) loaded per binding for evaluation; predicates may use context.<field>.
Policy — deterministic allow / require approval / deny before anything is queued for execution (optionally using context.* from loaded warehouse rows when bindings exist).
Approval — human in Slack/email when required.
Execution — webhook or API after the path clears.
Receipt — durable outcome in Snowflake views.

Where to look

APP.ACTION_DEFINITION / APP.ORG_ACTION
APP.ORG_POLICY
APP.GOVERNANCE_CONTEXT_SOURCE
APP.ACTION_CONTEXT_BINDING
APP.GOVERNANCE_AUTHORING_POLICY
APP.ORG_EXECUTOR_CONFIG
APP.RECEIPTS_VIEW

Governance authoring

Register context sources, bind them to actions, and compile policies into durable tables—via Snowflake procedures and/or the Tessra Action Service (including MCP tools). Narrative and operator flow:

Governance authoring & context →

Shape

CALL APP.REQUEST_ACTION_FROM_CORTEX(
  'ad_budget.increase',
  OBJECT_CONSTRUCT('amount', 500, 'currency', 'USD'),
  'Planning'
);
-- CALL APP.GET_ACTION_STATUS_FOR_CORTEX('<intent_id>');

API / SQL reference → · Quickstart →