Workbench
The AI-powered MDM surface for managing data quality workflows.
Workbench
The Workbench is the orchestrator inside the funnel — it sits between your sources (ingest) and your destinations (push), running the autoconfig → match → cluster → survivorship pipeline and tracking every step. It's an AI-powered surface for managing data quality workflows: surfacing issues, executing tool calls, and recording every change through an immutable event log.
Where the tool packages are the matching engine you can self-host, the Workbench is the hosted layer that drives them against your connected sources and hands the results to your destinations.
Core Concepts
| Concept | Description |
|---|---|
| Projects | Top-level containers scoped to a workspace, team, or organization |
| Notebooks | Event-sourced execution contexts that track every tool call |
| Chat | AI assistant that proposes and executes tool calls on your behalf |
| Trust | Gate system controlling how tool calls are approved before execution |
| Events | Immutable log of all actions with rewind and branch capabilities |
| Issues | Quality issues surfaced by tool execution, organized by severity |
Layout
The Workbench uses a three-zone layout designed for efficient data quality triage:
| Zone | Position | Width | Contents |
|---|---|---|---|
| Issues + Preview | Left panel | 60% | Issue list with detail preview |
| Chat | Right panel | 40% | AI assistant conversation |
| Event Log | Bottom drawer | 100% | Chronological event stream |
Tip: The bottom drawer can be collapsed when you want to focus on issues or chat. Expand it to inspect the full history of tool executions in the current notebook.
The left panel displays the issue list at the top and a detail preview below. Select any issue to see its full context, including the source event that produced it. The right panel hosts the AI chat assistant, where you describe goals and the assistant proposes tool calls to resolve them.