Clarity is a speech-practice app for iOS and Android. Read a passage, work through a one-minute drill, or speak off the cuff. Clarity follows along, scores the session, and turns the result into a concrete next step.

Website · Join the iOS beta

• Guided and freestyle practice. Use the built-in library, add your own passage, target one speaking skill with a drill, or answer a random prompt.
• Live feedback. A reference-aware teleprompter tracks the current word while showing elapsed time, pace, waveform, and transcript activity.
• Detailed results. Each session gets a speaking score out of 100 and five skill scores: Articulation, Flow, Pacing, Fillers, and Expression. Scripted sessions also include a word-by-word breakdown and recording playback.
• AI coaching. The result screen streams a short summary and three tips grounded in that session's measurements.
• Targeted follow-up. Clarity keeps a running list of difficult words, plays model pronunciation, and can generate a short passage that practices them.
• Progress tracking. Week, month, and all-time views cover score trends, practice time, sessions, streaks, skill movement, records, and mastered words.
• Account sync. Clerk handles sign-in. Convex syncs sessions, custom passages, and settings across devices without making the first screen wait on the network.

1. expo-speech-recognition streams native transcription while the app records the session audio.
2. Clarity's incremental aligner maps that transcript onto the reference text. It handles skipped, repeated, inserted, and partially spoken words while the teleprompter is moving.
3. When Azure Speech credentials are available, Clarity adds word, syllable, phoneme, fluency, and prosody detail. Without Azure, the local alignment still produces a complete fallback result.
4. The app writes the result to MMKV first. A single sync bridge sends local records to Convex after Clerk authenticates the account.
5. Expo Router API routes handle coaching, generated practice passages, and pronunciation audio without putting the AI Gateway key in the app bundle.

• Expo SDK 57, React Native 0.86, React 19, and Expo Router with typed routes
• expo-speech-recognition and expo-audio for recognition and recording
• Azure Speech pronunciation assessment for optional phoneme-level analysis
• Clerk for authentication and Convex for account data sync
• MMKV for synchronous, offline-first local storage
• Expo Router API routes, Vercel AI Gateway, and AI SDK 7 for generated features
• RevenueCat for Clarity Pro purchases and subscription state
• EAS Observe for startup, navigation, practice, auth, and error telemetry
• SF Pro Rounded, Hugeicons Pro, Reanimated, and Expo Glass Effect for the UI

Clarity requires a development build. Expo Go cannot load the native modules used by speech recognition, MMKV, and RevenueCat.

• Bun and the native toolchain for the platform you plan to run, either Xcode for iOS or Android Studio for Android
• A Hugeicons Pro token for the private icon packages
• Matching Clerk and Convex development projects
• A physical device if you need to test microphone and speech behavior

Create the ignored local environment file first. Bun reads it during dependency installation, including the HUGEICONS_TOKEN referenced by .npmrc.

cp .env.example .env.local
# Fill in HUGEICONS_TOKEN and the required development values.
bun install --frozen-lockfile

Start the Convex watcher. Its first run links or creates a development deployment and writes the deployment URL and identifier to .env.local.

bun run convex:dev

Each Convex deployment also needs the Frontend API URL for the matching Clerk instance. Set this on the backend, not in .env.local:

bunx convex env set CLERK_FRONTEND_API_URL 'https://your-instance.clerk.accounts.dev'

Leave bun run convex:dev running, then build the native app in another terminal:

bun run ios
# or
bun run android

.env.example documents the values normally set locally or in EAS. These are the groups that matter during setup:

Variables prefixed with EXPO_PUBLIC_ are compiled into the client bundle. Keep server credentials, including AI_GATEWAY_API_KEY, unprefixed.

The iOS Simulator does not provide reliable speech input. Set EXPO_PUBLIC_MOCK_PRACTICE=1 for deterministic passage and freestyle sessions, or use the EAS simulator profile. That profile also enables QA-only sign-in and history seeding while disabling Expo Updates so a shared preview update cannot replace the test bundle.

bunx eas-cli build --platform ios --profile simulator

Use a physical device to test microphone permissions, audio routing, transcription quality, Azure assessment, and recording playback.

app/              Native screens, navigation, session flow, and API routes
web/              Marketing-site router root and matching API routes
components/       Shared UI and feature components
hooks/            React state and session orchestration
services/         Recognition, storage, sync, purchases, and other side effects
lib/              Pure scoring, alignment, history, and statistics logic
constants/        Theme tokens, passages, drills, topics, goals, and vocabulary
convex/           Authenticated backend functions and schema
types/            Shared TypeScript models
scripts/          Bun logic suites and release support scripts
.eas/workflows/   Preview, TestFlight, feedback triage, and verified-fix workflows

Screens read synchronously from the local stores. components/convex-sync.tsx is the only app-data bridge to Convex, so network state never becomes a screen's source of truth. Pure modules in lib/ remain free of React Native imports and run directly in the Bun test scripts.

All visual tokens come from constants/theme.ts. Reusable design primitives live in components/ui/. Read AGENTS.md before changing UI or backend code; it records the project's design-system and Convex constraints.

APP_VARIANT gives development, preview, and production builds separate app names, identifiers, URL schemes, and iOS icons so they can be installed side by side. EAS has four build profiles:

• development creates an internal development client.
• preview creates an internal build on the preview update channel.
• simulator creates a deterministic iOS QA build against development services.
• production creates the store build and auto-increments its version.

The EAS workflow on main fingerprints the native layer, builds or repacks the iOS binary, and uploads it to TestFlight. The feedback workflows can turn TestFlight reports or GitHub issues into simulator-verified fix pull requests. See the workflow guide for setup and operating notes.

MIT. See LICENSE.