thinkering

A mobile app for adults' personal learning. You tell thinkering what you want to learn and why; it builds a pedagogically sound path of well-scoped goals and generates short, interactive activities each day — introduce something new, strengthen what you know, and put it to use.

Status: in beta. Intake, Today, the activity player, Path, History, Me, backup and sync, the landing site and anonymous analytics are built; what's left before TestFlight is in RELEASING.md. Not built yet: a share extension, reminders, widgets, dark mode and an Android polish pass.

Getting started

pnpm install
cp apps/mobile/.env.example apps/mobile/.env  # configure local app integrations as needed
pnpm verify            # typecheck + lint + tests — the gate for every PR
pnpm --filter @thinkering/mobile dev    # Expo dev server (press i for iOS simulator)
pnpm --filter @thinkering/web dev       # landing page at localhost:3000

Dev builds run in fixture AI mode: every generation is a recorded response, so the whole app works offline, at no cost, with no API key. Set EXPO_PUBLIC_AI_MODE=proxy for real generations. To start over, Me → Settings → Developer clears or reseeds the data.

The development guide — repo layout, rules, commands, how to change the database or a prompt — is AGENTS.md. It's written for coding agents and applies to everyone. To contribute, see CONTRIBUTING.md.

Principles

• Local-first. Your learning data lives on your device (SQLite). Supabase backup and sync is optional and off by default.
• Small, honest AI. Claude generates paths and activities through a thin metered proxy; prompts are versioned, inspectable code, not black boxes.
• Calm interface. Minimal text, clear daily structure (Next / Strengthen / Go further), no gamification noise.
• Grounded in learning science. Activities are built from a library of learning strategies with research behind them.

Docs

Doc	What it covers
00-overview	Vision, vocabulary, decisions
01-product-spec	How each part of the app behaves
02-architecture	Stack, local-first storage, sync, AI access, security
03-data-model	What each table is for, and the invariants
04-ai-pipeline	Every model call: trigger, model, contract, cost controls, failures
05-activity-format	The Activity Document format and renderer contract
06-library	The activity library (learning strategies)
07-design-system	Color, type, texture, components, voice
08-analytics-and-privacy	What's collected, what isn't, and the privacy disclosures
10-testing	What gets tested, and what deliberately doesn't

Stack

pnpm monorepo · Expo (iOS-first, Android + web capable) · expo-router · TypeScript strict · expo-sqlite + Drizzle ORM · Next.js on Vercel (landing + API proxy) · Claude API · Supabase (optional backup/sync, usage metering) · Featurebase (community feedback) · Resend (private feedback) · PostHog (anonymous, on by default, off with one toggle) · NativeWind + shared Tailwind tokens.

License

The code is licensed under AGPL-3.0 (see LICENSE).