Your formulas stay.
Everything else gets better.
Formualizer reads the actual formula graph from your .xlsx file — every SUMIFS, VLOOKUP, INDEX/MATCH, named range, and cross-sheet reference. We don’t convert. We don’t approximate. We execute your model as-is.
We read your workbook the way Excel does.
Drop your .xlsx. Formualizer’s parser extracts the complete formula AST — every cell’s formula, every dependency relationship, every named range. The system auto-detects structural regions: input areas, calculation blocks, output tables, parameter cells. You see a map of your workbook’s architecture before anything runs.
The parser handles Excel and OpenFormula dialects. 320+ built-in functions. Array formulas, structured references, implicit intersection. The dependency graph uses topological scheduling with cycle detection.
You tell us what goes in. We figure out what comes out.
The manifest is a contract between your workbook and the outside world. You define which cells accept new data (inputs) and which cells produce the results you care about (outputs). Everything in between — the formulas, the intermediate calculations, the cross-sheet references — stays untouched. We execute it exactly as Excel would.
Every output has a receipt.
Upload new data. The engine injects it into the mapped input cells, recalculates the entire dependency graph, and produces outputs. But here’s what’s different: every output cell carries a full trace — which inputs it depends on, which formulas were applied, what changed vs. the last run. An AI agent can read that trace and explain any result in plain English.
AI that explains, not guesses.
When a number changes between runs, the diff engine identifies exactly what moved and why. A sandboxed AI agent — armed with the formula trace, the cell diff, and the input changes — generates a plain-English explanation. It doesn’t summarize. It doesn’t infer. It reads the actual math and explains it.
North Region Q4 revenue increased by $94,300 vs Q3. Two factors: (1) Service revenue increased by $52,800 — hourly rate was raised from $72 to $80 effective October 1, applied across 2,310 billable hours. (2) Product units increased by 380 (3,020 → 3,400), adding $68,400, partially offset by higher returns (-$26,900 increase).
Open-source. Battle-tested. Permissively licensed.
- Arrow-backed columnar storage
- Incremental dependency graph with topological scheduling
- 320+ Excel-compatible functions
- Cycle detection, volatile function handling
- Rust core with Python and WASM bindings
- MCP server for LLM agent integration
- Stateless CLI for scripts and pipelines
- Auto-detected regions, table profiling, formula tracing
- Fork, edit, recalculate, diff workflows
- Token-efficient tool surface (discover, profile, extract)
import formualizer as fz
wb = fz.Workbook.open("forecast_model.xlsx")
eng = fz.Engine.from_workbook(wb)
wb.sheet("Actuals").set_value(14, 2, 3400)
result = eng.evaluate_cell("Summary", 2, 4)
print(result) # $847,200We’re currently working with healthcare organizations on physician compensation, board reporting, and RVU tracking. The engine handles any Excel model — if your team depends on a spreadsheet, we’d like to hear about it.
See it on your own workbook.
Any workbook. Any industry. If Excel can run it, Formualizer can make it auditable.
Book a demo