elixcee

Headless Excel workbook automation in Rust and Python: edit workbooks, recalculate supported formulas, and run data-processing VBA without Microsoft Excel. The core is Rust, with a Python API (PyO3), a standalone CLI, and an experimental @elixcee/xlsx JavaScript/WASM package.

Version: 1.0.5. See the changelog for versioned changes. The experimental JS package remains private and is not published. English | 日本語 | 中文

elixcee is a workbook automation runtime, not a VBA-only execution tool. Use the same workbook model for direct data edits, supported formula recalculation, and execution, diagnosis, or testing of supported data-processing VBA. It is not a replacement for the Excel desktop application: UI features such as charts, dialogs, and screen updates are skipped, modeled, or reported according to the operation.

Choose by workflow

Choose elixcee when the workflow needs one or more of these without a desktop Excel installation: read/edit/save .xlsx or .xlsm, calculate supported formulas, run and diagnose supported VBA, or test workbook behavior in CI. For a complete Excel desktop object model or full OOXML feature compatibility, check the documented support boundaries before adopting it.

Install

pip install elixcee

Pre-built CLI binaries are published on the GitHub Releases page.

For a source build:

python3 -m venv .venv
source .venv/bin/activate
pip install maturin
maturin develop

CLI

elixcee <file.bas>... <MacroName> [--file input.xlsx] [--sheet Sheet1]
                         [--output result.xlsx] [--json]
elixcee check <file.bas>... [--entry MacroName] [--json]
elixcee snapshot <workbook.xlsx|ods> [--json]
elixcee test-workbook fixture.toml [--json] [--seed N] [--case N]
elixcee diagnose <file.bas>... <MacroName> --file input.xlsx [--json]
elixcee diagnose-workbook fixture.toml [--json] [--seed N] [--case N] [--cases N]

The run command accepts standard VBA modules (.bas, .vbs, .txt) and exported class modules (.cls). Use Module.Sub for a qualified entry point in a multi-module run. For valid invocations, --json emits a result/error JSON object on stdout; see docs/agent-contract.md for the contract.

Python quick start

import elixcee

vm = elixcee.Vm()
vm.set_cell(1, 1, 10)          # coordinates are 1-based, like Excel
vm.run("""
Sub DoubleIt()
    Cells(1, 2).Value = Cells(1, 1).Value * 2
End Sub
""", "DoubleIt")
print(vm.get_cell(1, 2))       # 20

vm = elixcee.load_workbook("input.xlsx")
vm.set_cell(1, 1, 10)                              # edit a cell
vm.set_cell_formula(1, 2, "=A1*2")                # add a formula
vm.recalculate()                                  # recalculate formula cells
vm.run("Sub ProcessData()\n    Cells(1, 1).Value = 42\nEnd Sub", "ProcessData")
vm.save_workbook("output.xlsx")

# Cell/formula edits also support bounded undo/redo.
vm.set_cell(1, 1, 20)
vm.undo()

# Group a large edit into one undoable operation.
vm.begin_transaction()
vm.set_range("A1:C1000", [[row, row * 2, row * 3] for row in range(1, 1001)])
vm.commit_transaction()
vm.undo()  # restores the state before the whole transaction

# Optional reader resource controls (the cancellation check is cooperative).
cancel = elixcee.ReadCancellation()
vm = elixcee.load_workbook(
    "input.xlsx", max_work_units=100_000_000, timeout_ms=30_000, cancellation=cancel
)

The Python API provides headless formula calculation, cell/range editing, bounded undo/redo, formula evaluation, ranges, sorting, merges, hidden rows/columns, sheet management, styles, tables, data validation, AutoFilter, defined-name inspection, pandas export, and .xlsx/.xlsm/.ods workbook I/O. See elixcee.pyi for signatures and behavior. Vm.tables() and Vm.data_validations() return typed structural metadata projections; they do not evaluate calculated-column or validation formulas. For loaded XLSX/XLSM sheets, Vm.sheet_id(name) and Vm.sheet_name_for_id(sheet_id) expose stable source identities independently of tab order and rename operations; new or ODS sheets have no inferred ID.

For large XLSX/XLSM files, open_stream(path, sheet=None) yields rows without materializing the whole workbook. Set include_row_numbers=True to receive (row_number, values) tuples, or max_rows=N/max_row_bytes=N/max_columns=N to bound a read. Set timeout_ms=N to bound how long each next() waits for another row. create_stream(path) provides an append-only XLSX row writer. Set max_rows=N, max_columns=N, and/or max_pending_bytes=N to bound accepted output. The byte budget is cumulative until close, not retained RSS; this is not a constant-memory guarantee. See limits and Unreleased G1 changes. For a separate per-row and total-work budget, use create_stream_bounded(...).

Vm(on_msgbox="skip") is the default. Use on_msgbox="error" to make a MsgBox call raise an error. Set Vm(timeout_ms=N) or pass timeout_ms=N to run_macro to bound VBA execution time.

The read-only CLI snapshot accepts --max-work-units N, --timeout-ms N, and --cancel-file PATH. Creating the cancel-file while a read is in progress requests a cooperative stop; a blocking filesystem read cannot be forcibly interrupted by this mechanism. Repeated runs on one Vm reuse the parsed AST; vm.fork() creates an isolated batch copy. vm.snapshot() returns detached sheet values, tab order, defined names, calculation mode, visibility, merges, and hidden intervals. Set include_formulas=True for formula text. This is richer than the CLI snapshot schema. Use diagnose_macro(vba_code, macro_name, workbook_path) for structured diagnostics matching the CLI diagnose --json contract.

Supported VBA and formulas

The interpreter supports common data-processing constructs including Sub/Function, variables and arrays, If, For, For Each, Do, Select Case, With, On Error, user-defined types, named ranges, multiple sheets, Excel-style Range/Cells operations, and the documented built-in VBA Collection, in-memory Dictionary, and class-module subsets (see the documented limits). Formula support includes arithmetic, comparisons, criteria functions, lookup functions, date/time, text, statistical, financial, logical, and dynamic-array functions.

The maintained coverage list is FUNCTIONS.md. Unsupported or intentional no-op behavior is documented there and in the diagnostic contract.

Workbook compatibility

The Rust reader/writer preserves supported cell data, formulas, styles, merges, hidden rows/columns, and many unknown OOXML parts. Macro projects in .xlsm files are preserved during supported round trips. Features not modeled by the writer can still be lost or disconnected; tables, drawings, comments, hyperlinks, and other OOXML objects should be treated as compatibility gaps unless covered by tests for the version in use.

The project runs Rust tests, property tests, compatibility fixtures, and differential tests for the JavaScript package in CI. Compatibility with Excel's VBA execution semantics is not claimed for every macro, and post-save macro execution has not been fully validated against desktop Excel.

The v1 support boundaries are documented in docs/v1-support-contract.md. Version 1.0 is a stable contract for the documented data-processing subset and its safe failure behavior; it is not a claim of complete Excel or VBA compatibility.

Development

cargo test --workspace
cargo fmt --check
cargo clippy --all-targets --all-features -- -D warnings
cargo audit --no-fetch
cargo deny check --disable-fetch
python3 scripts/check-reader-measurements.py docs/measurements/*.json
python3 scripts/check-reader-measurements.py --self-test
bash scripts/check-measurement-boundary.sh

Offline dependency checks use the local advisory database and deny.toml. A stale database does not establish the absence of newly published advisories.

The short-term plan is in ROADMAP.md. Public design and policy documents are in docs/.

License: MIT licensing information. See third-party notices.