You do the thinking, the AI does the modelling, or both
Import Excel, write rows as functions, run the solver, diff the output — all from the CLI or through the agent loop.
Screenshot coming soon
Write Excel-like rows as Python functions. The framework builds the DAG, resolves dependencies, and solves in topological order — no graph wiring needed.
Round-trip any .xlsx file. The importer converts formula bands into clean Python rows; the exporter writes formulas back and verifies every cell matches.
NPV, IRR, XIRR, VLOOKUP, cumsum, running_max — the full Excel financial toolkit, callable from any row or scalar without leaving Python.
Claude iterates on your model: one change at a time, run → diff → commit. Skills keep it disciplined — bottom-up, override-aware, Excel-faithful.
Every solve is compared to the committed snapshot. Cell-level drift surfaces immediately. CI gates merges when numbers move unexpectedly.
Hardcoded adjustments live in overrides.toml — never baked into formulas. Every override carries a reason, author, and period.