DataFoundry

An enterprise-grade Data Agent workbench — understand business definitions through unified semantics, run multi-table and multi-step analysis inside a read-only safety boundary, make every step auditable and replayable, and turn one question into trusted data analysis.

![DataFoundry web workbench](assets/readme/gui-demo.gif){ .df-shot }
Web workbench, terminal workflow, and backend runtime share one governed task model.
### Semantic Accuracy Schema, metrics, field relationships, and datasource context are resolved before SQL is written, reducing guessed tables, wrong joins, and definition drift.
### Controlled Data Boundary Read-only SQL, credential isolation, masking, row limits, timeouts, and audit trails keep enterprise data work safer than a general coding agent.
### Data-task Runtime Multi-step analysis converges through schema inspection, tool policy, event streams, artifacts, and replayable history across Web, TUI, and API.
## One runtime, three entry points

Use the Web workbench for visual analysis review, the TUI for SSH and terminal workflows, and the API / AG-UI runtime when you need to embed the same governed task model into your own product.

TUI guide · Watch terminal demo

DataFoundry TUI demo

Start with the right path

### Try locally Run the workbench and ask your first question against the built-in DuckDB demo.

Open quick start{ .df-card-action }

### Understand the system Review the runtime, security boundary, and Data Gateway architecture.

Read architecture{ .df-card-action }

### Integrate APIs Use REST configuration APIs and the Agent Runtime / AG-UI entry point.

View API reference{ .df-card-action }