https://youtu.be/_emas5Vk2wQ?si=wT6_WyWjrVFxG6ax
English · 简体中文
Paste a prompt into Claude, answer a short interview, get a live financial model. Not a spreadsheet. Not an app. A single interactive artifact — model on the front, analysis on the back — built from your business, in your numbers, in a few minutes.
Revenue is never typed in. It's computed from what actually drives it. The balance sheet ties out every month. And a non-finance founder can read the whole thing.
One artifact, two sides — flip from the model to the analysis.
The model — every driver editable on the left; the whole 36-month model recomputes across ten sheets.
Flip the page over → a plain-English analysis: the one number that matters, red flags, benchmarks, and a bottom-line verdict.
The screenshots above are a real model generated by this prompt. Yours is built from your business, in your numbers.
Paste PROMPT.md into a fresh Claude, Claude Code, or Codex session and say:
"build my financial model"
It interviews you for a few minutes — what you sell, what drives revenue, your team, your costs, your funding — then generates a live interactive HTML artifact you can steer with sliders. That's the whole product. No install, no deploy, no database.
Prefer a slash command? Install it as a Claude Code skill and run /financial-model.
One self-contained HTML artifact with two sides:
The model (front)
- A headline KPI strip — Y3 revenue, break-even month, lowest cash, runway, LTV:CAC, burn multiple.
- An editable assumptions panel — every driver as a slider or number input, grouped and tagged GIVEN (you told us) vs ASSUMED (a preset filled the gap), each with a plain-English hover explainer. Change anything; the whole model recomputes instantly.
- Ten sheets as month-by-month tables: Overview · Revenue · Costs · Expenses · Team/Payroll · P&L · Cash flow · Balance sheet · Unit economics · Runway. Trend sparklines, in-cell magnitude bars, red negatives, emphasised subtotals.
- Scenarios — Base / Best / Worst, toggled live; the overview chart shows all three at once.
The analysis (flip the page over)
- A plain-English readout of your business model and growth engine.
- A red-flags section — the honest list a skeptic would raise, derived from your computed numbers and severity-tagged.
- A sector-benchmark check — your growth, churn, margin, LTV:CAC and payback against typical ranges, chip-scored.
- A bottom-line verdict and a "what has to be true" list.
Under all of it: a real three-statement engine — P&L, cash flow (indirect method), and a balance sheet that ties out to the penny every month. That tie-out is the model's built-in lie detector. If it balances, the accounting is internally consistent. If it doesn't, the model tells you it's wrong.
There's nothing to run. PROMPT.md is a prompt, not a program — it instructs the agent to interview you and then write a single HTML file that Claude renders as an Artifact. All the CSS and JS are inline; there are no external calls (Artifacts block them). The model lives entirely in that one file. Download it, share the link, or paste the prompt again tomorrow with different answers.
The engine is a single pure computeModel(assumptions) over 36 months. Every edit re-runs it from scratch — no stored state, no stale cells. That's why moving a slider updates all ten sheets at once.
flowchart TD
A["Paste PROMPT.md into Claude / Claude Code / Codex"] --> B["Agent interviews you:<br/>what drives your business?"]
B --> C{"Every driver answered?"}
C -->|some gaps| D["Sector presets fill the blanks<br/>(tagged ASSUMED, editable)"]
C -->|all given| E["Agent writes one<br/>self-contained HTML artifact"]
D --> E
E --> F["computeModel(assumptions)<br/>runs 36 months"]
F --> G["MODEL page<br/>10 sheets + KPI strip"]
F --> H["ANALYSIS page<br/>red flags, benchmarks, verdict"]
G --> I["You edit an assumption<br/>or drag a slider"]
I -->|instant recompute| F
Revenue is never typed in — it falls out of the drivers, then flows through the three statements, and the balance sheet has to tie out.
flowchart TD
subgraph IN["Drivers you control (inputs)"]
D1["Launch cohort"]
D2["Growth % / mo"]
D3["Churn % / mo"]
D4["Price / ARPU"]
D5["Team, salaries, opex, CAC"]
D6["Funding, working capital"]
end
IN --> CUST["Active customers<br/>= last month + new - churned"]
CUST --> REV["Revenue (an OUTPUT)<br/>= active customers x price"]
REV --> GP["less COGS<br/>= Gross profit"]
GP --> EBIT["less Payroll + Opex + CAC<br/>= EBITDA"]
EBIT --> NET["less Depreciation + Tax<br/>= Net income"]
NET --> PL["P&L statement"]
NET --> CF["Cash flow (indirect)"]
NET --> BS["Balance sheet"]
BS --> TIE{{"assets = liabilities + equity<br/>— ties out every month ✓"}}
Change any driver on the left and the whole chain re-runs; if the tie-out ever breaks, the model tells you the accounting is wrong.
Drop the skill into your Claude Code skills directory so you can invoke it by name:
# clone, then copy the skill into your Claude Code skills folder
git clone https://github.com/torlyai/financial-model-artifact
cp -r financial-model-artifact/skills/financial-model ~/.claude/skills/Then in any session:
/financial-model
The skill is a thin wrapper around PROMPT.md — same interview, same artifact.
This repo is the free scratchpad tier: one prompt, one artifact, ephemeral. Perfect for a first pass, a pitch-deck appendix, or a gut-check before you build.
When you want the model to persist — saved scenarios you can revisit, versioned assumptions, a team that can edit the same model, exports, and a real assessment rather than a self-drawn benchmark — that's TorlyAI. Same philosophy (revenue is an output, the balance sheet ties out), with the durability and rigour a funding conversation needs.
→ torly.ai/financial-model · 中文：torly.ai/zh/financial-model
Use this to think fast. Move to TorlyAI when the model needs to stick.
- Revenue is an output, never an input. Nobody knows their revenue — they know their drivers. Ask what moves the number (customers, price, growth, churn), then compute the number. A model you can type your hoped-for revenue into is a wish, not a model.
- The balance sheet must tie out. Every month,
assets = liabilities + equity. This isn't accounting pedantry — it's the one check that proves the three statements are consistent with each other. If the model can't tie, it can't be trusted. - Readable beats sophisticated. A founder who can't read their own model can't defend it. Plain labels, hover explainers, visible red flags, an analysis page that talks in English. Rigour underneath, clarity on top.
MIT. Free forever.
Built by TorlyAI. Agent-native, artifact-first financial modelling for founders and the people who build with them.