Finances Perso IA + Compta Indie
Dix picks pour le fondateur solo, freelance ou indie hacker qui tient ses propres comptes : OCR de reçus, catégorisation de dépenses, P&L simple, runway de trésorerie, multi-devises, et workflow d'estimation trimestrielle d'impôts. Plus léger que le pack comptable — pas de cérémonie d'audit, juste clôturer le mois en une heure.
What's in this pack
This is the stack for the person who is the company — a solo SaaS founder pulling Stripe payouts into a personal account, a freelance designer invoicing four clients across three currencies, an indie hacker with one product live on Lemon Squeezy and a day job slowly winding down. Not an accountant. Not an FP&A team. One person, one set of books, one hour a month to close them.
We already shipped a heavier Finance + Accountant AI Pack for the people who need an audit trail, plain-text ledgers, and Quant Analyst agents. This pack is deliberately lighter. No Hledger. No double-entry vocabulary. No Quant Analyst modeling. Just: paper in, categorized, on a P&L, runway visible, invoice out, tax estimate staged.
The operating principle: a solo operator's accounting needs to fit between two coffees on Sunday morning. Anything that takes longer doesn't get done, and the books rot. So every pick prioritizes friction reduction over auditor-grade rigor. If you grow past 1–2 people, graduate to the accountant pack.
Install in this order (ingestion → categorize → P&L → cashflow → tax)
- Zerox — Zero-Shot PDF OCR for AI Pipelines — start with the front door. Receipts from coffee shops, Uber, AWS, GitHub, Notion — they all arrive as PDFs or JPGs. Zerox uses vision models to extract structured rows in one shot. No template training, no learning curve. Pipe it into step 3.
- Marker — Convert PDF to Markdown with High Accuracy — Zerox is great for one-page receipts; Marker handles the long stuff: Stripe monthly statements, Mercury account exports, credit card PDFs with 4–6 pages and embedded tables. Markdown out, tables preserved.
- TaxHacker — Self-Hosted AI Accounting for Receipts and Invoices — the categorizer. Drop receipts in, get back categorized rows with multi-currency conversion built in. Custom prompts mean you can tune categories to your actual life: SaaS subscriptions, contractor payments, USDT off-ramps, whatever. Self-hosted so your card numbers never touch a third-party.
- Claude Official Skill: XLSX — Spreadsheet Operations — the bridge to a P&L sheet you can actually open. Once TaxHacker has classified the month, the XLSX skill reads/writes your monthly close workbook — appends rows, recomputes totals, keeps formatting. For an indie operator, this is your accounting system on month 1.
- Firefly III — Self-Hosted Personal Finance Manager — the upgrade path when the spreadsheet starts hurting. Firefly handles multi-currency natively (critical for anyone earning in USD/EUR/GBP), categorizes transactions with rules, and gives you an actual P&L view. Self-hosted on a $5 VPS.
- Actual Budget — Local-First Budgeting App — envelope budgeting for the founder side. Tells you
runway = cash on hand ÷ monthly burn, which is the only number that matters until you're profitable. Local-first means your books work on a plane. - Maybe — Modern Personal Finance OS — the visual layer. Net worth across accounts, multi-currency rollup, beautiful charts that make you look at your money instead of avoiding it. Pair with Firefly for the engine; Maybe is the dashboard your future self opens on a Sunday.
- Lago — Open-Source Usage-Based Billing — for indie SaaS revenue. If you're charging by API call, seat, or token, Lago is the open-source metering engine. Drops next to Stripe; revenue side of the books gets reliable, auditable numbers without a per-event SaaS fee.
- Invoice Ninja — Open Source Invoicing and Payments — the AR side for freelancers and consulting income. Multi-currency invoices, Stripe + bank transfer support, recurring billing, payment reminders. Self-hosted so client data stays yours.
- Dash — Data Apps and Dashboards in Pure Python — the always-on view. Build one dashboard: P&L year-to-date, cash runway, quarterly tax accrual (revenue × your estimated rate, ring-fenced in a separate account). Refresh nightly from Firefly's API. Replaces "let me run the numbers" forever.
How they fit together
Receipts / Stripe / Mercury PDFs
│
┌───────┴────────┐
▼ ▼
Zerox Marker
(single) (multi-page
statements)
│ │
└────────┬───────┘
▼
structured rows
│
▼
TaxHacker
(categorize + FX)
│
┌─────────┼──────────┐
▼ ▼ ▼
Claude XLSX Firefly III Lago
(sheet P&L) (self-host (usage
cashflow) billing)
│ │ │
│ Actual Budget │
│ (envelope/runway) │
│ │ │
│ Maybe (UI) │
│ │ │
└─────┬───────┴──────────┘
▼
Invoice Ninja ───── AR feeds back
│
▼
Dash
(live P&L + runway +
quarterly tax accrual)
The critical join is TaxHacker → Firefly III: once your transactions land categorized in a self-hosted hub, every downstream view (Maybe, Dash, your accountant at year-end) reads from one source. Avoid the indie trap of three half-finished spreadsheets and a Notion page nobody updates.
Tradeoffs you'll hit
- XLSX vs Firefly III — start with the Claude XLSX skill on month 1. You don't need Firefly until you have >100 transactions/month or are managing two currencies. Upgrading is one CSV export.
- Self-hosted vs SaaS — every tool here can self-host on a $5 droplet. The tradeoff is the half-day of Docker setup. If your time is worth >$200/hr, pay for QuickBooks Self-Employed and use this pack as the data-feed layer.
- Personal vs business books — keep them separate from week one, even if your business is unincorporated. The receipts pipeline is identical; the categorization rules diverge. Two TaxHacker instances, one Firefly with two budgets — done.
- AI categorization vs rules — TaxHacker will guess 90% of categories correctly. The other 10% will be silent miscategorizations. Spend 10 minutes a week reviewing the prior week's classifications. That weekly habit is what makes this pack work; skipping it is what makes books rot.
Common pitfalls
- Treating Stripe payouts as revenue — the payout net of fees is not your revenue line. Revenue is gross charges; Stripe fees are an expense. Lago (or a manual rule in Firefly) keeps these separate. Skipping this distinction breaks your tax estimate.
- Forgetting FX gain/loss — if you invoice in EUR and hold USD, the exchange-rate difference between invoice date and payment date is real money. Multi-currency tools track it; spreadsheets quietly lose it. Decide on cash-basis (record at payment FX rate) and document the rule.
- Quarterly tax estimate not ring-fenced — the tools tell you what you'll owe. They don't move money. Set a standing rule the day a payout lands: 25–30% (or your local equivalent) transfers to a separate "tax" account before any other spending. Dash is for visibility; the actual transfer is human discipline.
- Mixing in personal expenses without a flag — every solo founder has done it: paid for a personal Spotify on the business card. Don't fight it; flag it. Add a
personal-reimbursablecategory. At quarter-end, true it up. Two TaxHacker categories cost you ten minutes; an audit costs you a weekend. - This is editorial, not tax advice — every jurisdiction has its own rules on what counts as deductible, how quarterly estimates work, and when self-employment thresholds kick in. The tools make the workflow fast; they don't make you compliant. Talk to a local tax professional before any filing decision.
10 ressources prêtes à installer
Questions fréquentes
How is this different from the Finance + Accountant AI Pack?
Audience and audit ceremony. The accountant pack is for someone who has to close monthly books for a company with employees, vendors, and a board — it leans on Hledger (plain-text double-entry), Quant Analyst agents for FP&A modeling, and a Report Generator for board decks. This pack is for one person. No Hledger. No Quant Analyst. No double-entry vocabulary. The categorization layer (TaxHacker) and the OCR front door (Zerox, Marker) overlap, but the back end is self-hosted personal-finance apps (Firefly III, Actual, Maybe) instead of a ledger. If you ever hire your second employee, you'll graduate; until then this is the right amount of structure.
What's the smallest version I can start with this week?
Three picks: Zerox (OCR receipts), the Claude XLSX skill (write categorized rows into a sheet), and Actual Budget (envelope view so you can see runway). That's a working close pipeline in an afternoon. Add TaxHacker the first time manual categorization takes more than 20 minutes a month. Add Firefly III when the spreadsheet hits 200 rows or you're juggling two currencies. Add Invoice Ninja the first time you have to send a recurring invoice. Don't install everything at once — adoption pace matters more than completeness, and the pack's value compounds over months, not weekends.
Can I plug Stripe and Mercury directly into this?
Sort of — via exports, not live API. Most US indie operators pull a monthly PDF from Stripe and Mercury (or a CSV), drop it into Marker to extract the table, then pipe into TaxHacker for categorization. Live API integration exists for both (Firefly III has community Stripe importers; Mercury offers a read-only API), but the export workflow is more robust for a solo operator who closes monthly. Don't sink a weekend wiring live integrations until you have 12 months of clean monthly closes behind you. If you charge by usage rather than flat subscription, Lago sits next to Stripe and handles metering; that's worth wiring earlier.
Is multi-currency really first-class here?
Yes for the picks where it matters. Firefly III handles multi-currency natively (set a base currency, every account can be in a different one, FX rates fetched daily). Maybe rolls up across currencies for net worth. TaxHacker converts at the receipt's FX rate. Invoice Ninja invoices in any currency and tracks payment in another. The Claude XLSX skill is currency-agnostic — you'll build the conversion column yourself, which is fine on month 1 but painful on month 12 (that's the trigger to graduate to Firefly). The one place to be careful: cash-basis vs accrual-basis FX treatment is a real accounting choice with tax consequences. Pick one (cash-basis, recorded at payment date FX rate, is the indie default) and document the rule.
How do I stage a quarterly tax estimate without it becoming a full FP&A model?
Three-step ritual, not a model. First, the day every payout lands, transfer a fixed percentage to a separate savings account. The percentage is your blended effective rate from last year, or 25–30% if you're brand new (talk to a local pro about your jurisdiction). Second, on the last Sunday of each quarter, open the Dash dashboard: year-to-date revenue × your rate − amount already set aside = top-up needed. Third, file. That's it. The pack supports this ritual: Firefly tracks the revenue, Dash visualizes the accrual, TaxHacker keeps the categories clean enough that revenue is actually revenue. No DCF, no scenario analysis. The most common indie tax mistake is not running this once a quarter, and the tools fix that by making it 15 minutes instead of half a day.
12 packs · 80+ ressources sélectionnées
Découvrez tous les packs curatés sur la page d'accueil
Retour à tous les packs