MyMoney AI Disclosure
Effective Date: 2026-05-06 Last Updated: 2026-05-06
This document is a plain-English explanation of how MyMoney uses AI, what it sees, what model produces the output, and how to turn it all off.
1. Provider and model
MyMoney's AI features are powered by Anthropic's Claude API. Specifically:
- Claude Sonnet (current version:
claude-sonnet-4-6) — the default model for analytical features (advisor, recategorizer, anomaly detection, weekly digest, goal coach, cash flow forecast, rebalancing, tax-doc OCR). - Claude Haiku — used for fast, narrowly-scoped parsing tasks (e.g. Time Machine natural-language input in the Simulator).
Anthropic is a U.S.-based AI safety company. Their privacy and data usage terms are at anthropic.com/legal/privacy.
2. What AI does inside MyMoney
| Feature | What the AI does |
|---|---|
| Advisor chat | Answers your questions about your spending, budgets, goals, taxes, investments. Grounded in a snapshot of your real data. |
| Transaction recategorizer (F1) | When you correct a transaction's category, suggests a per-merchant rule and applies it across past + future transactions. Also a one-shot bulk pass over uncategorized rows. |
| Bill classifier (F2) | Pre-fills cadence, category, and amount range when you add a new bill. |
| Subscription audit (F3) | Monthly review of your recurring charges with cancel recommendations. |
| Anomaly detectors (F4a) | Flags unusual large transactions, category spikes, missing income. |
| Weekly digest (F4b) | Sunday-night narrative summary of the past week. |
| Goal coach (F5) | Weekly check-in commentary on each goal — behind, ahead, milestone, stalled. |
| Health score (F6) | Six-component financial health score with Sonnet narrative. |
| Receipt → transaction match (F7) | Deterministic with AI assist on the OCR step. |
| Pattern proposals (F8b) | Mines your correction log for proposed automations. |
| Cash flow forecast | Sonnet narrative on top of a deterministic 365-day projection. |
| Tax-doc OCR | Vision-enabled extraction of W-2, 1099, 1098, receipt fields. |
| Rebalancing analysis | 13-bucket asset allocation review against your target. |
| Tax history audit | Year-over-year longitudinal review of your tax returns. |
| Tax tips | Weekly Monday tips email (Sonnet). |
| Wizard chat | Budget + goal creation wizards with sanity-checked numerical proposals. |
| Simulator insights | Narrative on backtest / paper-trading results. |
3. What data is sent to the AI
For most features, we build a per-query "preamble" containing a slice of your data:
- Account balances (totals, not account numbers)
- Recent transactions: date, amount, merchant name, category, account
- Budget caps + spend-to-date
- Goal targets + progress
- Holdings: ticker symbols, quantities, market values
- Recent bills: name, amount, cadence, category
- Income totals (last 30/60/90 days)
- Net worth and cash-flow forecast inputs
We do not send:
- Your legal name, email, phone number, address, government ID, or other directly identifying personal information
- Your passwords, OAuth tokens, API keys, or any encrypted credential
- Your uploaded tax documents as raw bytes (only OCR features that explicitly ask the model to read a file send the image)
- Other users' data
For the tax-doc OCR feature specifically, we additionally send the image bytes of the document you uploaded so Claude's vision model can extract fields. The image is processed once per click and not retained at Anthropic beyond the request.
4. Anthropic's data handling
Per Anthropic's API policies:
- No training on your data. Anthropic does not train models on traffic submitted via the API by default. We use the standard API and do not opt into any training-eligible program.
- No retention beyond processing. Anthropic retains API request data only as long as needed to serve the response and meet operational/legal obligations (typically ~30 days for abuse monitoring), then deletes.
- No vendor sharing. Anthropic does not share your data with third parties for marketing.
We have a Data Processing Agreement (DPA) with Anthropic; see Privacy Policy section 11.
5. Accuracy and limits
AI output is informational and not professional advice. Specifically:
- Recommendations from the Advisor, Rebalancing, Tax tips, Tax history audit, Cash flow forecast, and Goal coach are not financial, tax, legal, or investment advice. We are not a registered investment advisor (RIA), broker-dealer, fiduciary, CPA, or tax preparer.
- The recategorizer can mislabel transactions. You can correct any AI-set category; corrections become rules for future transactions and you can delete those rules at any time (Settings → AI rules).
- The anomaly detectors can fire false positives. You can dismiss any warning; dismissals are remembered.
- OCR can misread documents. Always review extracted fields before applying them to your tax history.
You alone are responsible for decisions you make using MyMoney — including ones that incorporate AI-generated insights.
6. How to turn it off
You can disable AI processing across MyMoney at any time:
- Settings → Preferences → AI features → off
Effects when AI is off:
- Every AI-powered surface (advisor, forecast, rebalance card, simulator insights, recategorizer, etc.) is hidden in the navigation.
- No further data is sent to Anthropic from your account.
- Existing AI rules you created (per-merchant overrides, pattern proposals you accepted) keep applying — they're deterministic now. You can delete them in Settings → AI rules.
- Past AI outputs (advisor messages, weekly digests, subscription audits) remain readable until you delete them or until automatic retention purges them.
There is also an operator-level kill switch (AI_ENABLED=false env). When set, every Claude call is short-circuited globally regardless of per-user preference.
7. Rate limits and cost controls
To avoid runaway costs, MyMoney enforces:
- Per-user daily AI call cap — counts API calls across all features and rejects when exceeded with an inline 429 message in the UI.
- Advisor chat rate limit — 20 messages per hour per user.
- Operator daily cap — global ceiling across all users (
ANTHROPIC_DAILY_CALL_CAP, default 200).
These caps protect both us and you. If you hit one, the failing AI surface tells you so and recovers automatically.
8. Updates to this disclosure
We may update this disclosure as features evolve. Material changes (new features that send new categories of data, a model swap, a change in retention) trigger a re-acceptance prompt at next login.
9. Contact
Questions about how we use AI:
Andrew Walsh · Owner, MyMoney andrewdanawalsh@gmail.com