Quiet Ledger brings the invoice, the payment record and the conversation into one open-source workspace.
A client needs to agree on the work. An accountant needs to know what was actually recorded. Those conversations should not require sharing the whole ledger or moving financial details into a public chat.
What works today
- Reusable client invoice templates, date pickers, CAD/USD/KRW and other supported currencies, discounts, premiums and accepted-payment controls.
- Private checkout links, receipts, saved records and an invoice email flow.
- Passkey or Nostr account sign-in, encrypted server storage, a scoped API and MCP, and free self-hosting under AGPL-3.0-or-later.
- Marmot-encrypted invoice conversations. The demo offers separate two-member direct chats and an explicit three-member group, each with its own history.
Why Marmot
The merchant browser and AI-simulated client/accountant participants have separate Nostr identities and MLS state. Marmot TS interoperates with native Marmot MDK/wn-agent participants through a restricted Nostr-compatible relay. Riley and Jordan demonstrate how a client and accountant can discuss scope, totals and receipts without mixing direct-message histories. Ledger facts stay authoritative: an AI reply cannot mark an invoice paid or move funds.
Try it in under three minutes
- Open the demo link, choose Hackathon demo, and enter the public demo password: marmot.
- Create an invoice; try KRW, a discount and a premium.
- Start a conversation with Riley. Switch to Jordan to check the invoice, then choose Group to bring both in.
- Simulate payment, ask Jordan to check the receipt, and click the invoice to open its checkout and receipt.
- Open the invoice email preview. Saved demo invoices remain available in the same browser for 30 days.
The silent backup walkthrough is 2 minutes 47 seconds, with setup and waits shortened and genuine UI frames held for reading. Conversations use real Marmot encryption and actual AI inference; demo payments and receipts are explicitly simulated. Wallet/provider adapters require users‘ own connections, and funded production acceptance is not claimed. Gmail sending requires private sender setup; demo email is preview-only. White Noise phone joining and optional Marmot self-copy are not qualified or implemented.
Privacy boundaries
The storage operator can decrypt invoice records. Participating agents, their operator and their AI provider can read messages those agents decrypt. This is not zero-knowledge accounting. Browser messaging keys stay on the device; clearing site data loses access.
Source
GitWorkshop/ngit is the primary: https://gitworkshop.dev/npub17j3f7cuhcvjqnf7d80x6rt9hmkkhdguwg4l2th0h4qnhqketr63s6xakwv/git.shakespeare.diy/quiet-ledger The GitHub URL is an exact mirror. Built with AI coding assistance. Riley and Jordan are fictional people simulated with AI for this demo, standing in for a human client and accountant. Ordinary invoice accounts do not require built-in AI staff.