SpreadsheetsforAgents
Agents work directly with formulas, ranges, pivots, and charts. Every change is a Git-style commit — review, diff, merge, and roll back.
Run multiple agents alongside humans — Agents work. You decide.
Install Univer CLI skills: npx -y skills add dream-num/skills -g --all Open this sample workbook with Univer CLI: https://univer.ai/cli-assets/hello.univer.zip After opening it, show me: - the local viewer link - what I can explore next For help, the Univer CLI Discord is: https://discord.gg/nThHPupraR
BUILT FOR
any agent and stack
How it works
A simple three-step workflow for agent-human collaboration on spreadsheets.
Agent creates or updates the workbook.
Including formulas, formatting, charts, conditional formatting, and pivot tables.
Review every change like a pull request.
Each spreadsheet action becomes a structured mutation, making diffs clear and reviewable.
Approve, reject, merge, or roll back changes.
Stay in control without losing work or breaking the workbook.
Core Capabilities
Powered by Univer
13k+ GitHub stars. No Excel, LibreOffice, or cloud required. 500+ formulas and real formula execution, conditional formatting, data validation, permissions, charts, pivot tables.
Fully Local by Default
Run core workbook workflows entirely on your machine, with no network, no login, and no cloud dependency.
Shell-Native by Design
Work seamlessly with awk, grep, sed, jq, Python, and compose spreadsheet workflows directly in your terminal.
Human Review for Agent Changes
Agents create and update the workbook. Every change is structured and reviewable, like a pull request. Humans can comment, approve, reject, merge, or roll back safely.
Multi-Agent Collaboration
Assign each of your agents its own permissions, and let multiple agents work together on the same workbook.
Incremental Updates
Handle large datasets and complex spreadsheets by updating only what changed instead of rewriting the entire file.
Let AI handle real spreadsheet work
Pick an example close to your work. This page shows what to ask AI for and what Univer helps it deliver.
Each example has been validated with a real Univer workflow; the page shows the request and result, not execution detail.
Turn raw data into a weekly report
AI cleans the raw data, adds formulas and metrics, flags items that need human review, and exports a handoff-ready report file.
Use Univer to prepare this weekly report file: clean the raw data, add key metrics, create a Review sheet, and export a file I can send to teammates.
Input
Raw Data
Source data, blanks, and format issues.
Calculate
Metrics
Formula metrics and summary results.
Review
Review
Items that need human confirmation.
Handoff
Handoff
Confirmed result exported as .xlsx.
Clean raw data
Normalize fields, blanks, duplicates, and obvious format issues before analysis.
Add metrics and formulas
Generate summary metrics, calculated columns, and formula results that stay editable in the workbook.
Create a Review sheet
Collect items that need human confirmation in one place.
Deliver an .xlsx file
Export an .xlsx file that can be opened, shared, and edited further.
Local preview
Check formulas, formatting, and Review sheets before delivery or further edits.
Save and sync
Save the result as a version so teammates or other AIs can continue from it.
Roll back
If the result is not right, discard unconfirmed changes and return to an earlier version.
File handoff
When you need to share externally, export an .xlsx file that can be opened and edited.
Benchmark & Reliability
Validated against real-world spreadsheet benchmarks and continuously evaluated across diverse occupational tasks.
SpreadsheetBench
Real-world spreadsheet task accuracy
Accuracy (%) · pass@1
GDPval
real-world tasks
across 44 occupations
open tasks
including 75 spreadsheet-related cases
Spreadsheets for Agents.
Agents work directly with formulas, ranges, pivots, and charts. Every change is a Git-style commit — review, diff, merge, and roll back.
Run multiple agents alongside humans — Agents work. You decide.