Platform
Three surfaces, one connected workspace
Data Explorer, AI Agent, and Dashboard aren't three separate products glued together -- they share the same underlying query and permission model, so what the agent does and what you do by hand stay consistent.
Data Explorer
Browse schema and write SQL by hand, when you want to.
A schema browser and SQL editor, designed by looking closely at how the best data tools in this space handle the same problem -- then adding two things they don't: a Wiki tab for documenting tables in context, and a Data Lineage tab for tracing where a table's data actually comes from. The agent and the manual SQL editor share the same permission model -- there's no separate, more-permissive path for either one.
- Schema browser plus a full SQL editor for manual queries
- Wiki tab -- document what a table means, next to the table itself
- Data Lineage tab -- trace a table’s upstream sources
AI Agent
Three panes: conversation, Trace, and Canvas.
The conversation pane stays focused on the question and the answer. Every tool call the agent makes -- schema lookups, SQL, chart rendering -- streams into a separate Trace pane in real time, so you can watch the reasoning happen instead of waiting for a final summary. Canvas collects everything the session produced.
- Read-only SQL whitelist -- no DDL/DML, ever, with timeouts and scan limits
- Every tool call is logged, whether or not it produced a result
- define_metric lets the agent save a reusable aggregation instead of re-deriving the same number differently each time
Dashboard
JSON DSL, editable by you or by the agent.
Dashboards are structured data, not static exports -- the same dashboard the agent writes to during a session is the one you can open and edit by hand afterward. Variables and filters are first-class, not a workaround bolted on top.
- Agent can read and modify an existing dashboard’s structure directly, not just generate a one-off chart
- Variables and filters apply across the charts that reference them
- A chart produced by the agent can be saved into a new or existing dashboard
Underneath all three
Permissions the agent can't work around
- The agent runs under the same row/column permissions as the person who asked -- no elevated service-account shortcut
- Org → project permission model, with scope promotion requests for moving a resource from personal to shared
- Force-MFA enforcement at the application layer, since a shared-domain multi-org model can’t rely on login policy alone
See it on your own questions
Private beta -- request access and we'll follow up.