Ledgerly

Private, self-hosted personal finance — your data in your own Cloudflare account, never anyone else's.

Ledgerly is a full personal-finance dashboard that deploys to your Cloudflare account: transactions in your own D1 database, receipts and statements in your own R2 bucket, served by a Worker you control. No vendor sits between you and your money data. No telemetry. No sample data. No guessing.

Features

• Dashboard — net worth, income/spending, savings rate, cash-flow and category charts, all driven by a persisted period selector. Trends only render when real data supports them; nothing is ever invented.
• Transactions — search/filter, inline category and tag editing, manual entry, delete with confirmation.
• Imports that never guess — CSV statements with column-mapping preview, sign-convention confirmation for ambiguous exports, strict rejection of unparseable rows, and duplicate-proof fingerprinting across every import path (manual, CSV, documents, sync API).
• Recurring & subscription detection — a deterministic, unit-tested engine (median-interval classification, false-positive guards) that suggests; you confirm or ignore, it never auto-creates.
• Budgets, goals, rules, documents — monthly budgets with real spend, savings goals, plain-language categorization rules, and a document vault storing original receipt/statement bytes.
• 30 display currencies, a first-run setup wizard, mobile-friendly layouts, and a type-DELETE full data-erase that actually erases.

Screenshots show demo data added and removed during capture — a fresh install starts completely empty, by design.

Quick start (local)

git clone <this-repo> ledgerly && cd ledgerly
npm install
npm run dev

Open http://localhost:5173. Everything runs locally (Workers runtime with local D1/R2 via miniflare); data persists in .wrangler/state/. No Cloudflare account is needed — the only feature that requires one is the Workers AI receipt-extraction provider, which has no local simulator and is disconnected in local dev unless you opt in with LEDGERLY_REMOTE_BINDINGS=true after authenticating (npx wrangler login or CLOUDFLARE_API_TOKEN).

Deploy to your own Cloudflare account

1. npx wrangler login
2. npx wrangler d1 create ledgerly → paste the database_id into wrangler.jsonc
3. npx wrangler r2 bucket create ledgerly-documents
4. npm run deploy
5. Privacy step (required): the app has no login screen by design — put the deployed Worker behind Cloudflare Access (Zero Trust → Applications) so only you can reach it.

The free tiers of Workers, D1, and R2 are far more than a personal finance app needs.

AI document reading (optional)

Settings → AI lets Ledgerly suggest transactions from stored documents. Two shapes:

• Receipt extraction — one receipt or invoice becomes one suggested transaction with per-field confidence and a review form.
• PDF statement extraction — one bank/card statement becomes a whole batch of proposed rows in a triage table: select, edit, and import the rows you want; duplicates arrive pre-flagged and unchecked; rows the AI couldn't fully read can't be imported until you fill the gap; a statement too long to read completely says so instead of pretending. (Anthropic provider only — Workers AI models can't read PDFs yet. CSV statements still import deterministically and stay the recommended path.)

Two providers, off by default:

• Workers AI — processed by Workers AI inside your own Cloudflare account, never sent to a third-party vendor. Works automatically on deployed Workers; in local dev it needs Cloudflare auth plus LEDGERLY_REMOTE_BINDINGS=true (see vite.config.ts) — without that, everything else still runs fully locally.
• Anthropic (bring your own key) — higher accuracy; sends document images to Anthropic's API under your key. The key is stored in your database, write-only, and never displayed again. Works in local dev with no Cloudflare auth.

Extraction only ever suggests — you review and confirm every transaction, and the same duplicate detection applies as for any other import.

The mail-in feed — bank sync with no bank connection

Your bank already emails you transaction alerts, e-receipts, and statements. Route them to an address you control (Cloudflare Email Routing → your own Worker, or any mail pipe via the token-guarded HTTP door) and they become proposed transactions in seconds — no bank credentials, no aggregator, no Plaid, works with any bank that can send an email.

Security model, in one line: only allow-listed senders are accepted, nothing is imported without your review, and inbound email never triggers AI. The allowlist starts empty, so a fresh install ingests nothing. Parsing is deterministic — bank formats are community-contributed parser packs (code, not data). Setup guide: docs/EMAIL-FEED.md.

Chat with your money (MCP)

Ledgerly is a read-only MCP server: add your own instance to Claude Code, Claude Desktop, or any MCP client and ask questions in plain language — "what did I spend on dining last month?", "what's my 60-day cash flow?" Your client talks directly to your Worker; Ledgerly never calls any model, and no mutating tool exists. Setup: docs/MCP.md.

Scheduled imports

GET/POST /api/drive-sync implements a complete import contract (processed-file ledger, reset fencing, 20 MB limits, duplicate fingerprints) for any scheduled runner you authorize — set the SYNC_TOKEN secret (npx wrangler secret put SYNC_TOKEN) to require a bearer token.

Principles

1. Your data stays in your infrastructure. Default deployment is your own Cloudflare account; nothing phones home.
2. Deterministic money math, AI assist. Totals, dedupe, and detection are deterministic and tested. AI features suggest — they never silently write.
3. Never guess. Ambiguity asks; uncertainty goes to review.
4. Empty starts empty. No sample data, no fake trends, no dark patterns.

Roadmap

Receipt extraction and PDF statement extraction are shipped. Next up: categorization that learns from your corrections (repeated manual fixes become rules you approve), then cash-flow forecasting and a read-only MCP server so your own AI client can chat with your finances. The full direction is in docs/VISION.md, with the market research behind it in docs/research/.

Contributing

See CONTRIBUTING.md. The product spec lives in docs/SPEC.md; tests run with npm test.

License

AGPL-3.0. You can self-host, modify, and redistribute freely; if you offer a modified Ledgerly as a network service, you must share your changes under the same license.