I built an AI bookkeeping tool because I hated doing my own books
Hey everyone, long-time lurker here. After 4 months of nights and weekends, I finally shipped something I want to share — and more importantly, I'd love your honest feedback.
THE PROBLEM I COULDN'T IGNORE
I'm a freelance developer. I get paid through Stripe, PayPal, Wise, bank transfers, and occasionally Alipay when working with Chinese clients. Every month I'd download 4-5 different CSVs, each with different column layouts, date formats, decimal separators, and currencies. Reconciling them into a coherent picture took me 3-4 hours every month — usually on a Sunday I'd rather spend doing anything else.
I tried QuickBooks. It rejected my European-format CSVs. I tried Xero — it wanted me to configure a chart of accounts. I tried Wave, which was free but required me to manually categorize every single transaction.
The breaking point: a client sent me a PayPal payment screenshot. I had to manually type the amount, date, and currency into a spreadsheet. In 2026. I thought: AI should be doing this.
WHAT I BUILT
Tally Assistant — an AI-native bookkeeping tool for freelancers.
It does a few things:
• Dump any bank CSV and it works. Stripe, PayPal, Wise, Revolut, or your local bank. AI auto-detects columns, date formats, currencies, and delimiters. No column mapping, no reformatting. Supports 200+ bank formats.
• Snap a payment screenshot and AI extracts everything — merchant, amount, currency, date. OCR in 50+ languages.
• Describe an invoice in plain English and AI builds it. "Logo design $300, landing page $700 for Acme Corp, Net 15" becomes a complete invoice with line items, tax, and PayPal payment link as PDF.
• Auto payment reminders that escalate from friendly to firm over 21 days. No more awkward "hey, just checking on that invoice" emails.
• Multi-currency — 30+ currencies with daily exchange rates. Each transaction keeps its original currency, everything converts to your base currency for reporting.
Stack: Next.js 16 + PostgreSQL (Supabase) + Prisma + Clerk + OpenAI + Resend + Vercel.
Pricing: Free plan (50 transactions/month, 5 clients). Pro: $12/month. During launch, everything is free — I want people using it before I charge a cent.
WHAT I'M LOOKING FOR
I'm not here to sell you anything. I'm here because I want the IH community's honest take:
• Does the landing page communicate what this actually does?
• If you're a freelancer, would you use this? Why or why not?
• What's missing that would make you switch from whatever you're using now?
• Any obvious mistakes I'm not seeing?
Also: if anyone here does freelance work and wants to try it, I'll give you free Pro access for a year. Just DM me or comment, and I'll set you up. All I ask in return is honest feedback — the brutal kind.
Link: https://tallyassistant.com
Roast away. I can take it.
Love this, built straight out of real pain, which always shows. Handling messy bank CSVs plus OCR and invoice generation from plain text is genuinely useful, not just a gimmick. And your line about habits being the real barrier, not trust, is spot on, that's the harder problem most people don't admit. Good luck getting feedback from the Pro users!
The free Pro year will get you takers, but it selects for people who want free software over people who had your Sunday problem. The useful critics are freelancers already paying for something worse than what you built, and those you probably have to charge, even a few dollars, or you can't tell polite enthusiasm from need.
You asked about obvious mistakes. That's the one I'd watch, because it's the kind that shows up as good early feedback and bad month-three retention.
Which of the 200 bank formats did you actually test against a real statement, versus inferred from a spec?
This is a great example of solving your own pain point. The interesting part is that the problem isn't just bookkeeping itself — it's the friction of collecting data from multiple sources and turning messy financial information into something useful.
I think the biggest challenge with AI bookkeeping will be building trust. Freelancers may tolerate manual work, but they will be cautious about letting AI categorize transactions, generate invoices, or handle financial records without confidence in the accuracy.
Curious — how are you approaching the trust aspect? Are you showing users why a transaction was categorized a certain way, or allowing them to correct and teach the system over time?
The interesting part here is reducing bookkeeping from a recurring admin task into something freelancers can ignore.
Curious whether the biggest adoption barrier will be trust in AI handling financial data, or convincing freelancers to switch from spreadsheets and existing tools?
Great question. For me, the bigger barrier is habits, not trust.
Trust is easy to fix. I don't touch money, I don't store data for training, and I never ask for login credentials. It's read-only CSV parsing. Once people hear that, they relax.
Habits are the real fight. Freelancers have been using spreadsheets for years. They know it's painful, but it's familiar. Overcoming that inertia is harder than overcoming fear of AI.
That's why I made the first import instant. No column mapping, no reformatting. One CSV, two seconds, done. That moment is the switch. Once they feel the relief, they don't go back.
So trust is a conversation. Habits are a war of attrition. And I'm betting on the product experience to win the war.
Curious if you see it differently though. Open to being wrong here.
I appreciate you sharing how you're thinking about it. I do have a few thoughts, but they're a bit more nuanced than I'd want to squeeze into a thread.
What's the best email to reach you on?