Home · Quick start · Editor · charter.yaml · Sources · Agent · Guides · Evals · Audit · Policies · CLI · MCP · Workspace · Desktop · About · FAQ

Everything on this page works with no AI configured — the workbench is a complete local SQL tool on its own. datacharter serve, open the URL, explore.

The editor

A Monaco editor (the VS Code editor) with DuckDB SQL, catalog-aware autocomplete, and live results: a beat after you stop typing, the query runs automatically (row-capped, silent on error) so results track your typing. Cmd/Ctrl+Enter or Run executes the full query. Every run lands in a local history you can reopen, and Save keeps named queries in your workspace’s queries/ directory — plain .sql files, committed with the contract.

Results show their work

Under every result: the row count and Reads … — the exact source columns the query touched (provenance), with per-output-column lineage available via datacharter lineage. Flip Agent view to re-render the same result exactly as an agent would receive it — masked columns, row filters, policies applied.

Charts

The Chart tab auto-detects a chart from your result’s column types — dates + numbers become a line, categories + numbers a bar, two numbers a scatter — and captions it. Override the type (bar, line, area, scatter, pie) and the x/y columns from the dropdowns. Charts are Vega-Lite underneath; the agent can emit chart specs into the same panel.

Profile

The Profile tab summarizes every column of a result in one pass: nulls, distinct counts, quartiles, standard deviation, and per-column top-value bars. It’s the “what am I even looking at” button for an unfamiliar table.

Know the cost before you run

Estimate predicts how many rows a query will scan and warns before a big one; Explain shows the engine’s plan. Both live next to the Run button.

Files, snapshots, exports

Drag a CSV, Parquet, JSON, or Excel file onto the window and it’s queryable instantly (uploads are copied into the workspace and capped at 2 GB — for bigger files, add them as a source instead: that registers the file in place, no copy, no size limit, with pushdown). Snapshot saves a result as a reusable local.<name> table — each snapshot in the sidebar has a recheck button that re-runs its query and tells you inline whether the numbers changed. Export writes a result to CSV, Parquet, JSON, or XLSX — in Agent view, exports are masked too.

Command palette

Cmd/Ctrl+K jumps anywhere: tables, saved queries, panels, actions — and any contract-defined metric (“Run metric: revenue” compiles it to SQL and runs it). The fastest way around once you know it exists.

Next: The workspace on disk →