agami

The trust layer between AI and your database, so an agent's answers are defensible. Every join, metric, and named filter is confidence-scored and either FK-derived or human-approved via a review dashboard. Every answer ships a SQL receipt showing what ran and which model version it used. Works across Postgres / MySQL / Snowflake / BigQuery / Redshift / SQL Server / Oracle / Databricks / Trino / DuckDB / SQLite. Local-first: credentials, schema, results never leave your machine.