Documentation · Updated July 2026

From a spreadsheet problem to a result you can inspect and save.

No formulas, scripts, or JSON required. Describe the goal, inspect the constrained actions Tablyra proposes, and preview the finished result locally.

Quick start

Complete the first spreadsheet workflow in five steps.

  1. 01

    Install and sign in

    Download the Windows or macOS app and sign in with a verified account.

  2. 02

    Create a task and add files

    Add XLSX, XLSM, or CSV files. Desktop reads sheets, fields, and safety-relevant workbook features.

  3. 03

    Describe the result you want

    For example, format birthdays as YYYY-MM-DD, remove duplicate customer IDs, or join files by customer_id.

  4. 04

    Review and execute the plan

    Inspect the file, columns, and parameters used by every action, then build the preview locally.

  5. 05

    Inspect and save the result

    Review Source, Result, Diff, Issues, and the Action Timeline, then choose an output mode.

Continuous conversation

Every message refines the same result.

Tablyra sends only the latest instruction and the current validated plan to the planner, preventing prompt growth over a long conversation. After a preview, continue with requests such as “only change blank birthdays,” “then sort by customer ID,” or “undo the last step.”

  • The latest instruction does not carry an ever-growing transcript.
  • When a source changes, Desktop refreshes the schema and rebuilds the full preview.
  • Undo clears the last active operation without modifying the file.
Combining files

Make matching keys and exception policies explicit.

Add a primary and lookup file and choose the fields that represent the same record. Then select left, inner, or full join, decide whether duplicate keys are rejected or keep the first or last row, and choose output fields.

Read the complete Excel merge guide →
Output modes

Preview and write use the same result.

New file

The default. Write the result to a new XLSX or CSV file.

Workbook copy

Preserve supported workbook structure and write into a copied file.

Update source

Runs only after an explicit choice, backup creation, and output verification.

Preview only

Inspect the result without producing a file.

Safety boundaries

Detect workbook features that may not be preserved safely.

Macros, charts, pivots, external links, Power Query, and formula-sensitive structural changes affect which output modes are allowed. Tablyra blocks risky paths or requires a workbook copy instead of silently dropping content.

Read the complete security model →
Troubleshooting

Identify whether the problem is the file, plan, or output stage.

Source file missing
Restore the file to its original location or remove the old source and add the current file.
Preview has no changes
Check the fields and conditions used by each action, then inspect Issues and the Action Timeline.
Planner asks for clarification
Reply with the requested field, format, or matching condition; the current task continues.
Output fails
Keep the run ID and attach the privacy-filtered diagnostic package through Desktop feedback.
FAQ

Frequently asked questions

Why do I review samples?

The planner needs a small amount of context to understand column meaning. You can remove or edit samples, and full rows or files are not uploaded.

Why does the plan sometimes ask a clarification question?

Tablyra stops instead of guessing when headers are duplicated, join keys are unclear, a format has multiple meanings, or the workbook presents higher output risk.

Do I lose the result after closing the preview?

No. The current in-memory result reopens directly. After reopening a task, Desktop uses the saved validated plan to rebuild the preview locally.

How do I report an execution problem?

Open feedback in Desktop and attach the privacy-filtered diagnostic package. Run IDs, stage timings, and action results help investigation without attaching the full workbook.

Get started

See every spreadsheet change before you save it.

Download the desktop app, open an Excel or CSV file locally, and build a reviewable result through conversation.