AGILAB for Excel users

Excel users do not need a cluster pitch first. The useful first question is smaller:

Can I keep Excel as the familiar workbook interface, run a reproducible analysis outside the spreadsheet, and get a workbook plus evidence back?

AGILAB’s current answer is a lightweight bridge: workbook-shaped input, refresh-friendly CSV output, and explicit evidence. This is not a full Office add-in and it is not yet arbitrary workbook import from the web UI.

What ships now

The packaged excel_workbook_proof preview creates an Excel-shaped proof without adding Excel-specific dependencies:

uv --preview-features extra-build-dependencies run python src/agilab/examples/excel_workbook_proof/preview_excel_workbook_proof.py

It writes:

~/log/execute/excel_workbook_proof/input_sales_workbook.xlsx
~/log/execute/excel_workbook_proof/sales_proof_workbook.xlsx
~/log/execute/excel_workbook_proof/agilab_evidence.json
~/log/execute/excel_workbook_proof/power_query_refresh/sales_input.csv
~/log/execute/excel_workbook_proof/power_query_refresh/sales_summary.csv

Open sales_proof_workbook.xlsx in Excel and inspect the AGILAB Evidence sheet. The JSON evidence records artifact hashes and keeps the adoption boundary explicit.

Why this is the right first bridge

Excel analysts already understand workbooks, sheets, tables, refreshes, and folder-based data handoff. AGILAB should meet that workflow before asking them to think about workers, DAGs, Cython, or clusters.

The bridge keeps the responsibilities clear:

User concern

AGILAB response

I want a workbook I can open.

Produce sales_proof_workbook.xlsx with result and evidence sheets.

I want refreshable data.

Write stable CSV outputs under power_query_refresh/.

I need to know what changed.

Record SHA-256 hashes in agilab_evidence.json.

I do not want an Office deployment project.

Use files and folders first; keep add-ins as a later option.

Product direction

The next product step is a focused command and UI path:

agilab excel-proof --workbook sales.xlsx --sheet Sales --out sales_proof.xlsx

That should:

  • read a user-selected workbook sheet or table;

  • run a local AGILAB proof;

  • write a result workbook with an AGILAB Evidence sheet;

  • write Power Query-friendly CSV or parquet outputs;

  • keep a JSON evidence bundle with hashes, inputs, run metadata, and limits.

What remains roadmap

The preview does not yet claim:

  • arbitrary .xlsx upload/import in the AGILAB web UI;

  • formula preservation for existing customer workbooks;

  • Excel macro or .xlsm execution;

  • an Excel ribbon/task-pane add-in;

  • Microsoft 365 tenant deployment.

Those are useful later only if workbook import/export proves that Excel users care about the bridge. For now, the adoption message should stay simple:

Keep Excel as the interface. Use AGILAB to make the analysis reproducible, auditable, and replayable.