Financial modeling, as code.
Claude CodeClaude CoworkCodexGeminiCopilot edits your operating model the way it edits a software codebase.
Purpose-built MCP servers and CLI connectors.
Your agent reads, edits, and runs the spec from inside its native environment — not a chat box bolted onto a workbook.
- Claude Code
- Claude Cowork
- Codex
- Gemini
- Copilot
The leap is one step.
Every other AI-for-Excel tool is trying to teach an LLM to read a binary. We changed the substrate instead.
LLMs are extraordinary at code.
You've already seen it. Claude Code ships features in an afternoon. Cursor refactors entire repos. Models that struggled with spreadsheets a year ago now write production software.
So we made financial modeling code.
A clean DSL — JSON and markdown, the formats LLMs are fluent in. Not a black-box xlsx binary parsed character by character. The model is text, version-controlled, diffable, and reviewable.
And got an extraordinary modeler.
Claude Code reasons about your operating model the same way it reasons about a codebase. Drivers, schedules, scenarios — iterated in seconds. The LLM's full coding capability, pointed at financial work.
Built for IC, BvA, and ASC 820
Quarterly reporting, scenario stress tests, fair value workflows, IC memos — all running on a single JSON-spec operating model per portfolio company.
Drill from any number to its spec line and cell
Every AI answer cites the line in the JSON spec and the Excel cell behind it. Conversational drill-down on every driver. An auditable trail for ASC 820 fair value and IC defensibility.
Quarterly cycles, automated
BvA, actuals roll-forward, sensitivity refresh, AI-driven red-flag review — re-run on the latest spec when actuals land. The model is text, so cycles diff cleanly. No manual rebuild.
Portfolio-wide consistency
One JSON spec template per portfolio company, with a consistent assumption framework across the book. Run scenario stress tests across the whole portfolio in minutes.
Excel-backed. Cited answers. AI-native.
Two layers. Excel is the trust layer — the workbook your IC opens. A JSON spec is the modeling substrate — where Claude reads, edits, and iterates.
Excel: the trust layer
Calculations land in a real workbook your IC opens, audits, and hands off. Excel is the deliverable — not the substrate the AI has to fight through.
JSON: the modeling substrate
Drivers, schedules, scenarios — expressed in a JSON spec Claude reads and edits like code. The financial reasoning happens where the LLM is strongest.
Cited answers
Every AI answer traces back to a line in the spec and a cell in the workbook. No formula soup. No black boxes.
Compounding
Diffable, branchable, version-controlled. The model evolves like code — knowledge accrues cycle after cycle, instead of starting over in a chat window.
AI Quality Layer
Continuously running across every model, every cycle, every property above.
Insight
Conversational drill-down on every IC number — every answer cites the spec line and the cell behind it.
Speed
Portfolio-wide scenarios in minutes. Restructure in seconds — the model is already in the language Claude speaks.
Accuracy
Auditable trail for ASC 820 and IC defensibility. One assumption framework across the book.
Three outcomes, one DSL.
From IC decisions to quarterly cycles to LP communication — every workflow runs on the same modeling-as-code substrate.
Deal Evaluation
Iterate at the speed of code.
Rebuild a model on new diligence in minutes. Compare structures, stress assumptions, run sensitivities — the way Claude Code refactors a repo.
Learn moreOperating Decisions
Point an agent at the model.
Claude Code reasons about drivers, schedules, and scenarios directly. Test capital allocation, surface what needs partner attention, drill from any number to its source.
Learn moreQuarterly Reporting
Diffable models, every cycle.
BvA, actuals roll-forward, ASC 820 fair value — re-run on the latest spec. Version-controlled, auditable, agent-driven.
Learn more