Exporting your data

Three formats

Ztract exports extracted data in three formats. Pick the one your downstream tool reads natively.

JSON

The structured form. JSON preserves the full shape of your schema — nested objects, arrays of objects, arrays of strings — exactly as the engine returned them. Use JSON when:

• You’re piping data into another system via script or workflow.
• Your downstream tool reads JSON natively (most modern apps do).
• You care about nesting (parties on a contract, payment terms on a vendor agreement, line items with per-line tax).

CSV

The flat form. CSV is one row per document for the scalar fields, with line items emitted as a separate CSV when present. Use CSV when:

• You’re loading into a database, BI tool, or anything that expects flat tables.
• You’re working in command-line tooling (awk / cut / csvkit).
• You want a format that opens correctly in any spreadsheet without fiddling with delimiters.

Nested fields are flattened with dotted paths (e.g., parties_involved.party_1_name).

Excel

The spreadsheet form. Excel exports use multiple sheets when your schema has arrays — one sheet for the document-level scalar fields, one sheet per array (line items, transactions, etc.). Headers are human-readable. Use Excel when:

• The recipient is a non-technical user who’ll open the file directly.
• You need to combine extracted data with formulas before sending it on to someone else.

Where to export from

Open a project, switch to the Parsed data tab, and click Export. Pick the format, optionally narrow by parsed-date range, and confirm. Exports are project-scoped — the file contains every parsed document in that project that matches your date filter.

How exports are produced

Exports run as background jobs, not direct downloads. After you click Export, the job appears in the project’s Exports tab with one of three statuses:

• Queued — waiting for a worker to pick it up.
• Processing — the file is being built.
• Ready — done; click to download.

The dashboard polls the queue automatically, so you don’t have to refresh. For small projects the file is usually ready in seconds; for projects with thousands of documents it can take a few minutes. You can close the dashboard and come back later — the file stays available until you delete the project.

What ends up in each format

• JSON: an array of document objects, each matching the schema — nested objects, arrays of objects, and arrays of strings are preserved as-is.
• CSV: a flat layout with one row per parsed document (or per parsed page in per-page mode). Nested fields are dot-paths.
• Excel: one workbook with the same flat shape as CSV, formatted for direct opening.

Every export reflects your latest edits — if you corrected a value in the side-by-side viewer, the export uses your corrected value automatically (corrections are stored as overlays on top of the original extraction; nothing re-runs the engine).

Iterating on review and export

A common loop:

1. First-pass extraction (engine).
2. Spot-check confidence-flagged fields, correct as needed.
3. Run an export, use the file.
4. Notice a value you missed? Correct it. Re-export. Free — re-runs of the engine cost pages, but new export jobs do not.

API export

Programmatic export via HTTP API is on the roadmap. Today, all exports happen through the dashboard. Email [support@ztract.com](mailto:support@ztract.com?subject=API early access) if API access is a blocker and you want to be told as soon as it ships.