For advisors · AI-confident practices

Planning math your AI can't hallucinate.

A deterministic planning engine your assistant calls like a function — tax, projections, Social Security, equity comp, RMDs, cash flow, the whole plan. The AI does the modeling grind in seconds, so your time goes to the client, not the spreadsheet — and because another scenario costs almost nothing, you explore more of them and the plan comes out better. Every number still traces to the rule that produced it, so you can put it in front of a client or a compliance review and defend it.

Connect your assistant See how it works

At a glanceFor your practice

100%: Deterministic — same inputs, same outputs. Every figure traces to a calculation.
No SSN: No SSN field, no bank logins, nothing sold or tracked. Your clients aren't the product.
Any AI: Any MCP-compatible assistant — Claude, ChatGPT, Gemini CLI. It calls the engine; it never invents the math.

How it works

A calculator, not a guess.

FinPlan doesn't manage client relationships, replace your CRM, or generate PDFs. It does one thing exceptionally well: compute financial projections from structured inputs — and show its work for every number.

Step 1

Describe the scenario

Your assistant captures the client's accounts, balances, income, and filing status as plain structured data — no proprietary file format, no lock-in.

Step 2

Run the math

FinPlan computes the plan against the actual rules — IRS tables, contribution limits, phase-outs, AMT, RMD tables, Social Security regs — fast closed-form by default, full Monte Carlo when the path matters.

Step 3

Audit every number

Every output points back to the rule, bracket, or balance that produced it. You trust it because you can verify it — not because a vendor said so.

In a client review

Ask it the way you'd ask an analyst.

Your assistant turns each question into calls to the engine and reports what comes back, with the figures traced to their rules. Two are answered below on our sample household, the Larsons, straight from the engine.

1The client asks why their projection dropped since March. Diff the March snapshot against today's.
2Model exercising the ISOs and holding them against selling in the same year, and the tax each path triggers.
3How much should the Larsons convert to Roth this year to fill the 22% bracket, and what does it cost against leaving it in the IRA?

The Larsons · Roth conversion vs status quo · MFJ 2026 conversion income already taxed

Don't convert this year. Only $8,100 of the bracket is left, and keeping it in the IRA comes out $1,447 ahead after 30 years. Converting pays only if their rate in retirement tops 31.1%; we assumed 28%.

Taxable income · 22% bracket tops out at $211,400

$203,300 taxed +$8,100

After tax in 30 yrs at 6%

Convert now $32,049

Leave in IRA $33,496

Cost of converting $8,100 from the SEP-IRA

Federal tax at 22%
$1,782
New York tax at 6%
$486
Early-withdrawal penalty (under 59½)
$252

Break-even vs assumed rate in retirement
31.1% vs 28%

Question 3, run on the sample household (the Larsons): calculate_federal_tax_liability for their baseline, then analyze_roth_conversion with fill_to_marginal_rate: 0.22 and state_code: "NY". Wages, practice income and the SEP-IRA balance come from the profile; the 22% federal rate in retirement is our assumption.
4Compare claiming at 67 and at 70 for both spouses, including what the survivor keeps.

The Larsons · Social Security per month, in today's dollars both alive survivor keeps

Both claim at 67

Household $6,985

Survivor keeps $3,919

Both claim at 70

Household $8,662

Survivor keeps $4,860

Waiting to 70 raises the survivor's check from $3,919 to $4,860 a month, for life. For Mark alone, claiming at 70 pays more in total once he lives past 82.

Question 4, run on the sample household (the Larsons): estimate_social_security_pia_from_salary for each spouse ($3,919 and $3,066 a month from their profile income, over a 35-year career), then compare_social_security_claiming_ages once per claiming age with both spouses' records.

The whole plan

One engine, the whole plan.

Not a retirement-only silo. The same engine your assistant calls covers the questions a real plan actually raises — across every client, not just the ones near 65.

Tax

Federal and state income tax for all 50 states + DC, the local taxes that apply where they exist, capital gains, and AMT — against real IRS tables, the bracket math behind every projection.

Roth conversions

Size a conversion for this year: a fixed amount, fill to a bracket, or the largest amount worth converting, with the tax it triggers, the RMDs it removes, and the break-even against leaving it in the IRA.

Projections

Fast formula-based projections by default; full Monte Carlo when sequence risk or path-dependency matters — with client-ready chart data for every result.

Scenarios

Spin up what-if scenarios and compare them against a base plan — retire at 62 vs 65, saving more vs retiring later — side by side in seconds. Explore ten instead of one, and the plan you deliver is the one you actually pressure-tested.

Social Security

Claiming ages, breakeven, spousal and survivor benefits, and the earnings test — the math clients always ask about.

Equity comp

ISO exercises and the AMT that follows, RSU vests, NQSOs, and the disposition math — for the equity-heavy client others punt on.

Accounts & RMDs

Every common account type — retirement, education, taxable, property, and debts — plus RMD schedules and the tax cascade they trigger.

Cash flow & goals

Income and expense timelines, funding priorities, and goals scored by probability — the plan, not just the portfolio.

Why trust it

Auditable, composable, hallucination-free.

Snapshot it, then defend it

Freeze a plan into an immutable point-in-time record, and diff two snapshots to show exactly what changed between reviews. That's stronger evidence for a client or a compliance file than any screenshot — every figure still traces to the rule that produced it, and the whole engine is tested against those rules.

Immutable snapshots you can diff
Trace every figure to its rule
Tested against the real IRS tables

Compare scenarios, not just screens

FinPlan isn't a fixed set of tabs someone chose for you. Your assistant composes typed tools on the fly and stacks scenarios against a base plan — so the what-if a client raises mid-meeting gets answered on the spot, not in a follow-up email. Cheaper scenarios mean you test more of them, and a plan you've stress-tested from ten angles beats one you built once.

Compare scenarios against a base plan
Answer what-ifs live, not in a follow-up
Cheap scenarios → a better-tested plan

Your AI drives it — and can't fake it

FinPlan exposes a structured MCP interface. Plug it into Claude, ChatGPT, Gemini CLI, or any MCP-compatible assistant. The AI calls the engine like a function and reports what it returns — so the numbers come from the math, not the model.

MCP — works with any compatible assistant
Claude, ChatGPT, Gemini CLI
The AI calls the math; it never guesses it

Client data

One file per client, kept where your records live.

One file per client

Each client's plan is a plain JSON file. In local mode, FinPlan reads and writes it on your own disk, wherever you tell it to, and nothing is uploaded to FinPlan. Your AI provider still sees the conversation.

Kept under your controls

Store the files where client records already live: an encrypted drive, your document system, a git repository. Run local and we hold no client list and no client data. The hosted server keeps a plan while it's in use and for up to 60 minutes idle, then it expires.

Managed by hand, for now

There's no multi-client workspace yet. Opening a client means pointing your assistant at that client's file, and the files are yours to name and organize.

Due diligence

Check our work before you trust it.

FinPlan is financial planning software, not investment advice: it runs the calculations, and the recommendation stays yours. Everything below is public, so vet it the way you'd vet any vendor.

FinPlan is right for you if…

1You already use Claude or ChatGPT in your practice and want it wired to real math — not a chatbot bolted onto decade-old software.
2You have to explain and defend every number, to a client or to a compliance review.
3You've built your own spreadsheets because the planning software wouldn't show the work.
4You want to serve more clients — or go deeper for the ones you have — without your modeling time scaling with them.
5You'd rather a tool that never asks for a client's SSN or a bank login than one that stockpiles their data.
6You'd rather trust-and-verify a transparent engine than accept a recommendation you can't reproduce.
7You want one engine across the whole plan — tax, cash flow, equity comp, Social Security, retirement — not a retirement-only silo.

See it on your own hardest case.

Connect your assistant, load a scenario, and trace every number yourself — no sales call, and no client file to upload. Start on sample data, then point it at the real thing.

Connect your assistant Try it with sample data