Conversational planning
Understand goals such as normalizing birth dates, joining by customer ID, or validating an import using only the current instruction, schema, and approved samples.
Not AI changing a spreadsheet blindly. Every change is explainable, previewable, and reversible.
Tablyra separates natural language, constrained actions, local execution, and safe output so non-technical users get automation speed without giving up human review.
Four-layer operating model
AI understands the goal. Deterministic code handles the file.
Constrained action engine
The model can select only defined Semantic Actions. Column references, parameters, and output policies must validate before execution.
Local execution and preview
Desktop reads the file and builds Source, Result, Diff, Issues, and an Action Timeline without remote code execution.
Verified output
Preview and final output use the same change set. Updating a source file requires a backup followed by output verification.
Supported operations
Cover cleanup, formatting, structure, multi-file work, and data quality.
These capabilities run through explicit Action Schemas rather than arbitrary model-generated code.
Cleanup
Whitespace cleanup, text replacement, case changes, character removal, blank filling, and duplicate handling
Formatting
Date, phone, number, currency, percent, and display formats
Structure
Rename, delete, split, merge, select, and reorder columns
Rows
Filter, keep, delete, sort, blank-row, and duplicate-row operations
Multiple files
Join sources by explicit keys, select output fields, and handle duplicate keys
Validation
Required, unique, email, phone, date, number, range, and allowed-value checks
Product boundaries
Know when automation is safe and when it should stop.
CapabilityTablyraGeneric chat assistant
Full-file processingRuns locallyOften requires uploading or copying data
Operation scopeConstrained and validatedMay generate arbitrary formulas or code
Preview before saveSource, Result, Diff, and IssuesDepends on manual inspection
Source protectionNew file by default; backup before replacementDepends on the generated script
Iterative editsUpdates one plan and previewOften regenerates an answer
FAQ
Frequently asked questions
How is Tablyra different from asking AI to write a Python script?
Tablyra does not execute arbitrary model-generated code. Requests are constrained to defined and validated spreadsheet actions with visible results, diffs, and issues for every step.
Can I keep refining the same result through conversation?
Yes. Each message updates the current validated plan and local preview. A saved result can also be rebuilt after closing the preview or reopening the task.
Can it work with multiple sheets or files?
You can add multiple sources and join two files through explicit matching keys. Complex workbook features are detected and output is restricted when they cannot be preserved safely.
Which file formats are supported?
The current focus is CSV, XLSX, and XLSM. Macros, charts, pivots, external links, Power Query, and formula-sensitive structures are detected and handled according to output risk.
Get started
Turn the next spreadsheet cleanup into a repeatable, reviewable workflow.
Download the desktop app, open an Excel or CSV file locally, and build a reviewable result through conversation.