Invoicing tools (the usual accounting apps, payment-link makers and freelancer platforms) all start with a blank form, so the freelancer retypes a deal they already made in a chat. Chat to Invoice starts from the negotiation itself: it reads the thread, finds what was finally agreed (not the first price someone floated), and writes the invoice, the payment link and the reminder schedule in one step. Marketplaces do this only for jobs that happen inside their walls. This works for the deals people really make, in WhatsApp, Telegram, Instagram DMs and email.
The data it runs on
Source: chats the freelancer forwards. Three ways in: share a WhatsApp chat export (the app's own "Export chat" gives a .txt file), forward messages to a bot number, or paste text or screenshots.
Access: nothing is scraped. Only what the user shares is read, once, then the raw text is deleted and only the extracted fields plus quoted source lines are kept.
Limits and gaps: voice notes need transcription first; screenshots need OCR; chats mixing languages or slang ("400k", "half now") need careful parsing; some deals have no clear final price.
Fallback: when a field is unclear the AI asks one question ("Was the 420k with or without the interior shots?") instead of guessing, and every field stays editable before sending.
1. The mission
Make getting paid the easiest part of freelancing anywhere in the world. A deal agreed in a chat should become a professional invoice, a way to pay that works for the client, and polite follow-ups, without a single form.
2. Why people can't resist it
Instant relief: the dreaded admin task is done in ten seconds, from something you already have.
No awkwardness: the app chases money for you in a kind voice, so you never have to sound needy.
The payment ping: a big, satisfying "Paid" moment and a dashboard that grows every month.
3. How it works for real
Ingest: parse WhatsApp .txt exports (timestamps, senders), Telegram JSON, pasted text. Screenshots go through a vision model (Gemini 2.5 Flash via OpenRouter) for OCR. Voice notes through a speech model (Whisper class).
Extract: a frontier LLM (Claude Sonnet or GPT class via OpenRouter) returns strict JSON: parties, deliverables, quantities, final price, currency, deposit, dates, conditions, each with the quoted message it came from. A fast cheap model (Gemini Flash or DeepSeek) double checks numbers and flags conflicts.
Invoice: rendered HTML to PDF with a sequential number, tax fields per country (VAT, NIF, RUT where needed).
Pay link: card via a payment processor, local bank transfer via a regional rail or virtual account provider, and USDC on Base. For USDC the freelancer has a smart wallet made at email login (Coinbase Smart Wallet, Privy or Dynamic), gas sponsored by a paymaster, so crypto stays invisible. Cash out to local bank through an off-ramp partner.
Reminders: a scheduler sends messages written by the LLM in the client's language and the chosen tone, through WhatsApp Business API or email, and stops the second a payment webhook or on-chain transfer lands.
4. Cost per use
Pennies per invoice. Extraction about $0.01 to $0.03 with a frontier model, $0.002 for the check model, reminders about $0.001 each, PDF rendering near zero. USDC transfers on Base cost well under a cent and are sponsored.
Free: 5 invoices a month, all payment methods, reminders by email. Pro ($6 a month): unlimited invoices, WhatsApp reminders, your logo, multi-currency income dashboard, tax export. Card and bank payments carry the processor fee; USDC has none, which nudges clients toward the cheapest rail without forcing it.
5. Growth loop
Every invoice and payment page carries a small "Made from a chat" line. Clients who are freelancers themselves try it. Freelancers share the "paid in 2 hours" card in their communities. Referral: a free Pro month for both.
6. Risks and guardrails
Privacy: chats can contain personal matters. Process once, delete raw text, keep only quoted deal lines. Never train on users' chats.
Wrong numbers: every field shows its source quote and must be confirmed before the invoice goes out.
Harassment: reminders are capped, polite by design, and stop on request from the client.
Compliance: payment partners handle KYC; the app shows tax fields but does not give tax advice.
7. MVP build order
Paste a chat, get extracted JSON with source quotes, edit, render invoice PDF.
Hosted payment page with card and USDC on Base (email smart wallet, sponsored gas).