The most technically interesting engine of the four and the least ready for someone else's workbook. Opening your own .xlsx and recomputing it is plausible. Opening the finance team's file with PivotTables, Power Query and a decade of formatting assumptions is not — and the project says so first.
Two numbers that look like a contradiction
The README says GridCraft covers 89% of Excel's ribbon commands and implements 513 of 523 worksheet functions. The roadmap says it is about 50% ready for real work. Both are true, and the distance between them is the best short explanation of what "clone" means in 2026.
The function count is a measurement of the library, not of the work. The roadmap's own detail is more precise than the headline: 504 of the 523 functions actually compute, which is 96.4%. Nineteen are stubs or wrong. And breadth weighted by what people actually use — dialogs, formats, charts, the parts that are not functions at all — comes out at about 75%.
What actually works
- Native .xlsx read and write, plus CSV, with XLSB and ODS as data import.
- A real calculation engine — dependency graph, evaluator, dynamic arrays — not a formula string formatter.
- Copy-on-write cells, styles, tables, charts in the model layer, with a toolkit-free chart renderer.
- Fill, sort, filter, clipboard and history as commands, which is what makes the CLI and MCP mode possible.
- A swappable front end: the egui UI sits on top of a separate engine crate, and the project enforces the layering with
cargo xtask layers.
Signed builders exist for Windows, macOS, Linux, BSD and the web. An x64/arm64/x86 MSI, or three cargo words if you build it yourself.
What doesn't work yet
Confirmed working
- Core end-to-end spreadsheet workflow
- 513 functions implemented; 504 verified computing
- Native .xlsx round-trip
- Dynamic arrays, sort, filter
- MCP server + CLI
Documented gaps
- Get & Transform and the data model — absent
- PivotTables — not yet
- Very large workbooks; multi-threaded recalc
- XLSX fidelity on real-world files
- Localisation: 12 languages still queued
The roadmap's ranked work list is explicit about what dominates the remaining hours: file exchange with Excel users (real corpus, pass-through, large files, .xls, encryption), common-chart depth, PivotTables, keyboard and interaction, and calculation correctness. It estimates 160–240 hours to beta and roughly 700–1,100 to full parity.
The honest read on "fidelity"
A spreadsheet is not a document: it is a calculation with a presentation layer. Errors in the calculation are silent and catastrophic; errors in the presentation are cosmetic. GridCraft's own gaps list puts recalculation correctness near the top, which is the right priority — and the reason "513 functions" should not be read as "your model will compute the same".
Practical test before trusting it: open a workbook you know the answer to, force a full recalc, and compare. The project openly expects differences on exotic functions and on very large models.
Should you install it?
Try it if you want a fast native spreadsheet, you write your own formulas rather than inherit them, or you want an agent to drive a workbook through MCP.
Wait if your work depends on PivotTables, Power Query, or opening files from colleagues and returning them unchanged. Those are exactly the items on the roadmap, not in the build.
Sources: github.com/storytold/gridcraft (README, ROADMAP.md, docs/function-parity.md, docs/gaps.md), read 18 October 2026.