All projects
ios · swift · app · productivity · fintech · privacy

Kestra

A private personal dashboard for iPhone — money, health, training, spending and time on one screen. Local-first, no accounts, no tracking.

Kestra icon

What Kestra Is

Kestra is a native SwiftUI app that replaces the dozen apps you check each morning with one screen you build yourself. Money, health, training, food, spending, parcels, trains, calendar and weather — arranged into pages from a library of over forty widgets.

The thesis is that most screen time is fragmentation. Ten apps, ten feeds, ten sets of notifications. Kestra collapses them into a single glanceable surface with no feed and nothing to scroll — you check it, get what you need, and put the phone down.

Built Local-First

The architectural decision that shapes everything else: there is no server.

No accounts, no analytics, no tracking, and no backend the developer operates. Data lives in UserDefaults on the device; API keys live in the iOS Keychain. When a widget reads from Trading 212 or WHOOP, the request goes straight from the phone to that service using a key the user supplies. Nothing passes through a middleman, because there is no middleman.

That makes the App Store privacy label an honest “Data Not Collected” rather than a carefully-worded one. It also means zero hosting cost and no scaling problem — the app works identically for one user or a million. The trade-off is real and deliberate: no analytics means no visibility into how it’s used, and bring-your-own-key narrows the audience to people willing to paste an API key.

An optional personal server — one the user owns — adds Obsidian sync, the AI briefing and WHOOP OAuth. The app is fully functional without it.

The Widget System

Everything is a widget, and users assemble their own app:

Features Worth Calling Out

The Logger. Press the iPhone Action Button, speak, and Kestra works out what you meant. “Spent £20 at Tesco” files to spending and deducts from the right pot. “I ate porridge for breakfast” becomes macros. “Did legs, RPE 8” becomes a training session. “I have an idea…” goes to notes, and on to Obsidian. Routing is rules-based rather than a model call — deterministic, instant, works offline, and testable.

Earnings Radar. A nightly job pulls official filings from the SEC’s EDGAR system, scores fundamentals in Python, and writes a briefing on the three strongest companies reporting soon — with an analyst you can question. Every figure is computed deterministically; the language model only writes prose over numbers already verified, so it cannot invent a statistic.

Spending by voice, mapped. Purchases logged by speech are geotagged and plotted, so spending patterns show up geographically. The parser handles UK money slang — “a fiver”, “couple of tenners”, “two bands”.

Chess. A rewritten engine with full legal move generation, castling, en passant, promotion, piece-square tables and quiescence search, running off the main thread. Pieces are drawn as geometry rather than Unicode glyphs, so they match the rest of the app.

Design

A deliberately anti-default visual language: Instrument Serif for display type, Geist Mono for every number and label, warm off-white on near-black, and an acid-lime accent. Zero corner radius throughout — hard edges, hairline rules, data rendered as texture rather than chrome.

Personalisation is extensive: 18 themes, 24 accent colours or any custom colour, four typography styles, three layout densities and a bounded text-size scale.

How It’s Built

Status

Feature-complete and in daily use, preparing for App Store submission. Free to use, with an optional £4.99 one-off unlock for the features that cost real money to run.

Built solo, in Swift.