FZ-021 · ACCOUNTING
Ledgers
Open-source bookkeeping for businesses that let AI agents draft the entries.
- Status
- Alpha
- Pipeline stage
- VALIDATION
- Sector
- ACCOUNTING
- Autonomy target
- LEVEL 4 · SELF-OPTIMIZING
- Built with
- 3 live · 0 planned
The problem
Bookkeeping means matching invoices to orders and payments, and nobody wants an AI posting to the books unchecked.
What it does
Ledgers is designed as open-source accounting where agents draft entries and a person approves every payment out.
How it works
- Every agent key gets a tier (read, draft, post within limits) and amount limits per currency; paying out, closing periods and filing tax always need a person
- A dry run shows every debit and credit before anything posts, and every action is audited and reversible
- Agents use the CLI and MCP tools, people review in a browser and phone app, and all of them run the same commands
Planned to be free and open source (MIT) for a single user with your own AI model key; managed plans would include the agents.
Stop typing invoices into your books and only look at the few entries an agent isn’t sure about.
TodayWaitlist open at ledgers.sh; the app runs on staging for the owner; the code is public on GitHub; nothing is published to crates.io yet.
The film
Aims
- Autonomous operation
- The bookkeeping is the work meant to run without people: agents read incoming bills, match them to orders, receipts and payments, and draft and post entries within the limits each agent key is given. People stay in the loop by design: anything above a limit stops for approval, and paying money out, closing a period and filing tax always need a person.
- Intelligence
- A language model reads documents and proposes matches, each field with a confidence score and a link to its place on the source; the ledger itself is plain double-entry code. Every action is audited, entries are reversed rather than edited, and a dry run shows each debit and credit before anything posts.
- Growth
- Open source first: free for a single user with their own AI model key, so developers and their agents can start from the CLI and MCP tools. Growth is companies moving to managed plans, which include the agents, counted from the first document an agent books that is not ours.
The technical detail
Open-source, double-entry accounting built for AI agents. Agents extract vendor bills, propose invoice, order, receipt and payment matches with confidence scores, and draft journal entries; each agent key has a tier (read, draft, post within limits) and per-currency amount limits, and paying money out, closing a period and filing tax always need a person. Every action is audited and reversible, and a dry run shows each debit and credit before anything posts. Agents work through a CLI and MCP tools, people supervise in a browser and phone app (PWA, offline, push approvals), and an HTTP API shares the same commands. Multi-currency and right-to-left from the start; one company, internal units or a group with consolidation. Eight modules (ledger, invoice, bills, matching, bank, tax, documents, reports) are planned as Rust crates on the Cratefield harness, MIT-licensed, self-hosted or managed. What exists today is the homepage, the app on the owner’s staging environment and the public code on GitHub; nothing is published to crates.io yet.
