Natural-language agent

Reads data immediately and can propose changes — but every statement it produces is re-classified, re-authorized, and (for writes/schema changes) held for your confirmation by the backend, not by the model, regardless of which model answered.

Model DDL / ADMIN — choose a model for this turn, or use the configured default.

Review by another model

SQL console

Runs exactly what you type through the same classify → authorize → execute pipeline as the agent. SELECT/SHOW run immediately; everything else becomes a pending action.

Pending actions

Approval inbox for SQL, Dashboard and ETL database changes. Confirm or cancel actions here, or directly under the agent message that created them.

Status

No pending actions.

Business / architecture rules

Rule text is data, not code — every create/update/deactivate/delete goes through the same propose → confirm pattern as a schema change, and is audited either way.

Service

Propose a change

change_type

rule_id (blank for CREATE)

agent_type

rule_key (optional)

priority

title

rule_text

Agent models ADMIN

Only admins can change which model answers by default, or pick a one-off model for a chat turn. Everyone else always gets whatever's set as default here.

Audit log

Every request, decision, proposal, confirmation, cancellation, expiry, and denial is logged here — including the ones the model tried and the backend refused.

target user_id

event_type

start_time

end_time

limit

AI Dashboard DASHBOARD AGENT

Create a dashboard from scratch in Dashboard Chat, or pin a chart created elsewhere. Dashboard SQL is READ-only and always re-authorized by the backend.

Current dashboard

Dashboard Chat

Model Uses /api/dashboard/chat, not SQL Chat.

Visualization contract

Backend-enabled now: bar, stacked_bar, horizontal_bar, line, area, pie, donut, scatter, combo, kpi, table, and geo. Chart objects include the authorized SQL text for demo transparency.

AI ETL GUIDED ETL MVP

Upload a source file, profile the source, choose a compatible template after upload or build a new configuration, review mapping and target column definitions, then validate and load.

Create or open an ETL job.

ETL Chat

Model Uses /api/etl/chat.

Guided actions

These controls are rendered from ui_actions in the backend DataFlow JSON.

DataFlow JSON

Demo Database READ ONLY

Browse the same demo schema and data used by the SQL, Dashboard, and ETL agents. No inline edits are allowed here.

Tables

Not loaded.

Select a table

Choose a table from the left.