Cadence - Automatic Time Tracker

Gallery

About

Cadence is a finished, installable desktop app that tracks how you actually spend your working day — and coaches you on it with an AI model running on your own computer.It is built as a local-first alternative to Rize.io ($16.99/mo) and RescueTime ($12/mo). Same core loop — automatic tracking, focus scoring, AI insights — with one structural difference that turns out to be the whole pitch: nothing leaves the machine. No backend, no account, no telemetry, no per-seat cost. The database is a single SQLite file in the user's home folder, and the AI is Google's Gemma running through Ollama on their own CPU or GPU.That makes the economics unusual for this category. There is no server to run and no marginal cost per user — a buyer can sell it as a one-time licence, ship it as a paid Setapp/Gumroad app, hand it to an agency as an internal tool, or white-label it for a market where cloud time-tracking is a compliance problem: law firms, healthcare, defence, finance, anyone under GDPR scrutiny.What it does- Tracks automatically. Polls the foreground window every 4 seconds — app, window title, active browser URL. Stops after 30 seconds idle and instantly on sleep, screen lock or screensaver, so a closed laptop never bills you eight phantom hours.- Categorises itself. 61 built-in keyword rules handle the obvious cases for free; anything unknown goes to Gemma, whose answer is written back as a new rule. The AI teaches the rule engine, so the app gets faster and cheaper the longer it runs.- Scores focus, not just hours. A 0–100 Focus Quality Score weighing productive ratio (50%), low distraction (20%), context-switching depth (20%) and break balance (10%) — plus daily/weekly goals and a streak counter.- Detects sessions on its own. Sustained deep work opens a focus session; an idle gap opens a break. Manual focus/break/meeting sessions come with goals, timers and built-in focus music synthesised in the Web Audio API.- Blocks distractions. During a focus session, a distracting app triggers either a soft notification or a hard always-on-top overlay with take a break / dismiss.- Bills clients. Clients with hourly rates, keyword attribution rules, AI attribution for whatever the rules miss — including a generated work description and a learned keyword — inline reassignment in the timeline, earnings for today/7d/30d, CSV export.- Coaches. A daily insight and a chat coach, both grounded in the user's real database rather than generic productivity advice. The conversation is stored and restored between sessions.- Swaps its own brain. Settings lists every model installed in Ollama and can download a new one — qwen3, deepseek-r1, llama3.2 — with a progress bar, no terminal. The Ollama endpoint is editable too, so the model can run on another machine on the network.- Speaks English and Italian, switchable in Settings; the AI answers in whichever is selected.- Lives in the background. Menu-bar app on macOS, system tray on Windows, live totals in the tray menu, starts at login, single-instance locked, onboarding wizard included.TechElectron 33 · electron-vite · React 19 · TypeScript · Tailwind 3 · Recharts · better-sqlite3 · get-windows · Ollama/Gemma. Roughly 7,300 lines across a clean main/preload/renderer split: thirteen focused modules in the main process, one typed IPC bridge, eight React views, and a two-language dictionary with a consistency check that fails the build on an untranslated string.What the buyer receives- The complete, commented source — MIT licensed, no third-party licence to inherit.- Two-double-click installers, in separate macOS/ and Windows/ folders. Script 1 installs Ollama and pulls the Gemma model; script 2 installs Node if needed, builds the app, installs it and enables start-at-login. Each folder has its own README with requirements, permissions, file locations, uninstall and a troubleshooting table.- An illustrated 14-page user guide as a print-ready PDF, in English and Italian — every screen, the score formula, the categorisation pipeline, backup and troubleshooting. Rebuilt from HTML with one command, so it stays current as the product changes.- A brand kit: logo in SVG, PNG at every size, .ico, .icns, and 16 app screenshots at 2× — plus the one-command regenerators for all of it (npm run icons, npm run guide).- A demo-data seeder that fills a throwaway database with a realistic week, for demos, screenshots and support.- A file-by-file developer guide and the original roadmap, including everything that was planned and not built.Cadence is a finished, installable desktop app that tracks how you actually spend your working day — and coaches you on it with an AI model running on your own computer. It is built as a local-first alternative to Rize.io ($16.99/mo) and RescueTime ($12/mo). Same core loop — automatic tracking, focus scoring, AI insights — with one structural difference that turns out to be the whole pitch: nothing leaves the machine. No backend, no account, no telemetry, no per-seat cost. The database is a single SQLite file in the user's home folder, and the AI is Google's Gemma running through Ollama on their own CPU or GPU. That makes the economics unusual for this category. There is no server to run and no marginal cost per user — a buyer can sell it as a one-time licence, ship it as a paid Setapp/Gumroad app, hand it to an agency as an internal tool, or white-label it for a market where cloud time-tracking is a compliance problem: law firms, healthcare, defence, finance, anyone under GDPR scrutiny. What it does - Tracks automatically. Polls the foreground window every 4 seconds — app, window title, active browser URL. Stops after 30 seconds idle and instantly on sleep, screen lock or screensaver, so a closed laptop never bills you eight phantom hours. - Categorises itself. 61 built-in keyword rules handle the obvious cases for free; anything unknown goes to Gemma, whose answer is written back as a new rule. The AI teaches the rule engine, so the app gets faster and cheaper the longer it runs. - Scores focus, not just hours. A 0–100 Focus Quality Score weighing productive ratio (50%), low distraction (20%), context-switching depth (20%) and break balance (10%) — plus daily/weekly goals and a streak counter. - Detects sessions on its own. Sustained deep work opens a focus session; an idle gap opens a break. Manual focus/break/meeting sessions come with goals, timers and built-in focus music synthesised in the Web Audio API. - Blocks distractions. During a focus session, a distracting app triggers either a soft notification or a hard always-on-top overlay with take a break / dismiss. - Bills clients. Clients with hourly rates, keyword attribution rules, AI attribution for whatever the rules miss — including a generated work description and a learned keyword — inline reassignment in the timeline, earnings for today/7d/30d, CSV export. - Coaches. A daily insight and a chat coach, both grounded in the user's real database rather than generic productivity advice. The conversation is stored and restored between sessions. - Swaps its own brain. Settings lists every model installed in Ollama and can download a new one — qwen3, deepseek-r1, llama3.2 — with a progress bar, no terminal. The Ollama endpoint is editable too, so the model can run on another machine on the network. - Speaks English and Italian, switchable in Settings; the AI answers in whichever is selected. - Lives in the background. Menu-bar app on macOS, system tray on Windows, live totals in the tray menu, starts at login, single-instance locked, onboarding wizard included. Tech Electron 33 · electron-vite · React 19 · TypeScript · Tailwind 3 · Recharts · better-sqlite3 · get-windows · Ollama/Gemma. Roughly 7,300 lines across a clean main/preload/renderer split: thirteen focused modules in the main process, one typed IPC bridge, eight React views, and a two-language dictionary with a consistency check that fails the build on an untranslated string. What the buyer receives