Pocket Cash

Overview

Pocket Cash is an offline-first personal finance app for the desktop. It turns bank CSV exports into an organised view of spending, net worth and financial-year reporting, and it runs entirely on your own machine. There is no sign-in, no server, and no analytics. The database lives in the app's own folder and nothing leaves the device.

Key Features

• Bank-agnostic CSV import: Map a statement's columns once, and that mapping becomes a reusable importer for every future export from that bank
• Safe re-imports: Each account fingerprints its rows, so re-importing the same file skips duplicates instead of double-counting
• Categorisation: Organise transactions into categories and tags, with spending broken down per category and per month
• Net worth across accounts: Balances and trends aggregated across every account you track
• Tax-time reporting: Financial-year summaries and a tax report that exports to CSV
• Optional AI: Auto-categorising, insights and a deduction scan, off by default, and able to run fully offline against a local model

Technical Stack

• Shell: Electron, packaged as a downloadable desktop release
• UI: Next.js, React, Tailwind CSS
• Database: PGlite, Postgres compiled to WASM, running in-process
• Data access: Drizzle ORM
• Monorepo: Turborepo

Architecture

Pocket Cash inverts the usual finance-app architecture: instead of a hosted backend holding the data, the entire stack ships to the user's machine.

• Local database: PGlite gives a real Postgres engine with no server to run and no connection to configure
• Typed data layer: Drizzle ORM over that local database
• Desktop shell: Electron wraps the Next.js UI and owns the app's data directory
• Import pipeline: Column mapping, row fingerprinting and deduplication sit between a raw CSV and the ledger

Challenges & Solutions

• Privacy without a backend: Chose a local-first design so there is no account system, no hosting and no telemetry to secure in the first place; the privacy guarantee comes from the architecture rather than from policy
• Every bank exports differently: Solved with a reusable per-account column mapping rather than trying to detect formats automatically
• Re-importing overlapping statements: Row-level fingerprinting makes imports idempotent, so an overlapping date range cannot double-count
• Postgres with no server: PGlite provides real Postgres semantics in-process, avoiding both a bundled database server and the limitations of SQLite for reporting queries

Impact & Results

• Privacy by construction: Financial data never leaves the device, because there is nowhere for it to go
• Reusable importers: The one-time column mapping removes the repetitive part of tracking spending
• Tax preparation: Financial-year and deduction reporting exports directly to CSV
• Distribution: Shipped as a versioned desktop release with CI, under a PolyForm Noncommercial licence