Inspiration
I once won a hackathon and got paid through PayPal. It was an international payment. Instead of seeing the money right away, PayPal held it for about a month and asked me to explain where it came from. I did explain — and I still waited.
That frustration stuck with me. Freelancers, creators, contractors, and anyone paid across borders hit the same wall: the money arrives, then it freezes in review. People don’t know what documents to send, how to word the reply, or how the sender should have structured the payout in the first place.
Clearhold started from that moment: if AI and PayPal can move money, they should also help get stuck money moving again.
What it does
Clearhold is an AI co-pilot for held PayPal payments.
For recipients (Unstick funds) Paste a hold notice or case text. Clearhold classifies the payment, estimates risk, builds a document checklist, drafts a PayPal support reply, and outlines an invoice/source-of-funds pack.
For senders (Send safely) Create a documented invoice package and send a PayPal payout with an explicit purpose note, then track status — so the transfer has a clear trail before money moves.
There’s a Live / Demo toggle: Live uses free Gemini AI and PayPal sandbox APIs; Demo uses predictable offline responses for recording videos.
How we built it
• Frontend: Next.js (App Router), TypeScript, Tailwind • AI: Vercel AI SDK + Google Gemini (free tier) for structured hold diagnosis, with a local demo engine as fallback • Payments: PayPal REST APIs — OAuth, Invoicing, Payouts, and status — against the sandbox • Design: Unslopped UI toward a calm “payment paperwork / clearing desk” look: quiet chrome, warm neutrals, and a status case as the main visual • Deploy: GitHub + Vercel at clearhold-sigma.vercel.app
Challenges we ran into
• “Free AI” isn’t always free out of the box. Vercel AI Gateway free models still required a card on the team, so we switched to Google Gemini’s free API. • Model choice mattered. A heavier Gemini flash model was so slow the diagnose API timed out; we moved to a fast lite model. • PayPal holds can’t be bypassed. We had to position Clearhold honestly: reduce holds and resolve them faster with documentation — not magically release funds. • Vercel project URL. clearhold.vercel.app was taken, so production landed on clearhold-sigma.vercel.app. • Environment resets mid-build. We rebuilt the app, pushed to GitHub, and rewired secrets on Vercel.
Accomplishments that we’re proud of
• A working end-to-end product: diagnose → checklist → reply draft → invoice → sandbox payout → status • Real Live mode with Free AI and PayPal sandbox, plus Demo mode for clean demos • A design that feels like a payment case file, not a generic AI template • Shipping a public repo and a live deployment people can actually try
What we learned
• Cross-border PayPal holds are as much a documentation and communication problem as a payments API problem • AI is most useful when it produces something people can send today (a reply draft + checklist), not just a vague explanation • Sandbox + demo fallbacks make demos reliable when APIs or keys misbehave • Free models need careful picking for latency, or the product feels broken in production
What’s next for Clearhold
• Connect real (non-sandbox) PayPal for verified businesses who want production payouts • Deeper case tracking: save hold cases, documents, and outcomes over time • Better sender templates for freelancers, prizes, contractors, and marketplaces • Optional custom domain and polished mobile UX • Measure what actually shortens hold time — then double down on those flows
Built With
- css
- next.js
- tailwind
- typescript
