Playground · XLCRACK

Operate the cleanup pipeline, not just a chat window.

This workspace is wired to live backend routes. Every major action here is a tool-backed step with visible outputs and versioned results.

What To Run First

Sign up once or switch user id
Upload dataset and click Analyze dataset
Run preview prompt and inspect risk flags
Request approval and run recipe
Download latest CSV or save workstream
DatasetLoading rowslive
UploadCSV · TXT · XLSX
Drop a file or browse

CSV, TXT, TSV, or Excel

Guided Templates0 recommendations

Keep UI simple: detect the best stream from your real file, apply, then run guided preview/approval.

No recommendations yet

Upload/select a dataset and click detect to create agentic template options.

Saved Workstreams0 streams

Save modular crack streams per user, recognize known templates on new datasets, and rerun.

Data contextlive schema
Null revenue
Distinct dates
Distinct regions
ColumnTypeNulls
Loading schema…
Sample rowstool: sample_rows
Recipe buildertool: propose_recipe
No recipe yet

Run a preview to generate steps.

waiting
LLM prompt
Validationtool: validate_recipe
No validation warnings yet.
Preview & approvaltool: preview_recipe
MetricBeforeAfter
Rows
Row delta0%
No approval required
ExportSQL Server
Dataset: none
Tool call log0 events
Waiting for tool calls…

Storyboard the transformation

Every frame is a tool call with an auditable output.

Frame 01

Ingest & fingerprint

Capture raw file state, schema, and row counts.

get_schema · profile_columns
Frame 02

Draft recipe

LLM proposes structured steps with risk flags.

propose_recipe
Frame 03

Preview & validate

See diff impact and require approval.

validate_recipe · preview_recipe
Frame 04

Run & version

Execute in a durable workflow. New version every run.

run_recipe