PocketBrain v0.6.0

PocketBrain is a private, phone-first, local-first AI brain that runs in the browser, stores memory locally, and supports optional user-controlled bridge mode for stronger models.

Built by Michael Hughes / PHI369 Labs.

What's new in v0.6.0

• Lazy-loaded WebLLM and chunked bundles for faster initial startup.
• Camera/image memory UX improvements and timeline support.
• Device sync UX upgrades including QR pairing and manual sync controls.
• Built-in TTS controls with browser fallback behavior.

Built by Michael Hughes / PHI369 Labs.

What's new in v0.6.0

• Lazy-loaded WebLLM and chunked bundles for faster initial startup.
• Camera/image memory UX improvements and timeline support.
• Device sync UX upgrades including QR pairing and manual sync controls.
• Built-in TTS controls with browser fallback behavior.

🚀 Try PocketBrain

Launch PocketBrain - Runs entirely in your browser. No signup, no tracking.

Want to self-host? Clone the repo and run npm ci && npm run dev

Portfolio Highlights

• Local-first browser AI app
• IndexedDB private memory loop
• WebLLM local provider support
• Optional Ollama/OpenAI-compatible bridge mode
• Encrypted backup/restore with Web Crypto
• PWA install support
• TypeScript/Vite production build

Core product behavior

• Local-first by default: WebLLM runs in-browser, memory stays in IndexedDB.
• Private memory loop: pinned summary + recency + retrieval over local messages.
• Optional bridge mode: send compiled prompts only to endpoints you configure.
• Voice notes: capture audio locally; use browser speech recognition when available.
• Trusted-device sync foundation: pairing payloads + trusted bridge endpoint history.
• Encrypted backup/restore: PBKDF2 + AES-GCM in browser.
• PWA + GitHub Pages deployment support.

Modes

1) Local WebLLM mode (default)

• Provider: local-webllm
• No bridge endpoint needed
• Memory/retrieval stay on device

2) Bridge mode (optional)

• Providers: ollama-bridge, openai-compatible-bridge
• Endpoint, model, and key are user-configured
• Retrieval remains local; only final prompt is sent to the bridge
• Fallback-to-local is supported when bridge init fails

Voice notes and transcription

• Microphone capture always stores voice note locally.
• If browser speech recognition exists, transcript is inserted into draft.
• If not available, note is still saved and shown as non-transcribed.
• Bridge transcription is explicit/manual and provider-dependent.

TTS

• The app currently focuses on text responses and voice input.
• Browser/provider TTS playback is not yet fully integrated into the default UX.

Image capture and image memory indexing

• Image-memory retrieval hooks are now present in retrieval/search contracts.
• Full camera capture + persistent image memory UX is still in-progress.

Trusted devices / sync foundation

• Pairing-first bridge workflow includes:
  • trusted endpoint history
  • pairing payload/file import/export
  • explicit known-host probe (no blind subnet scans)
• This is a foundation for future trusted-device sync experiences.

Encrypted backup and restore

• Export plain or encrypted backup packages.
• Encrypted packages use browser Web Crypto (PBKDF2 + AES-GCM).
• Import supports merge/replace mode.
• Wrong passphrase fails safely.

Local setup

npm ci
npm run dev

Verification / release checks

npm run lint
npm run build
npm run verify

npm run verify is a lightweight CI-safe integration check that catches contract drift between retrieval/provider interfaces.

GitHub Pages deployment

1. Push to main
2. Enable Pages source = GitHub Actions
3. Actions build and deploy dist

Current limitations

• Large WebLLM bundles produce large PWA assets; initial load can be heavy on low-end phones.
• Bridge transcription and bridge image-description support depend on endpoint model capability and API compatibility.
• Full camera-to-memory workflow is not yet complete in the UI.

Privacy

No hidden telemetry. Bridge mode sends prompts only to endpoints you explicitly configure.

License

MIT

Getting Started

1. Web: Visit https://michaelwave369.github.io/PocketBrain
2. Mobile: Open the link on your phone and tap "Install" to add to home screen (PWA)
3. Self-hosted: Clone this repo and run:

npm ci
npm run dev

First run: Click "Get started" to configure your AI brain (local or bridge mode)

Quick Tour

• Record voice notes - Capture thoughts hands-free
• Local memory - Everything stays on your device in IndexedDB
• Ask your brain - Query what you've learned (uses WebLLM by default)
• Optional bridge - Connect to Ollama or OpenAI for stronger models
• Encrypted backup - Export and restore your memories safely

Why PocketBrain?

• Private by default - No telemetry, no tracking, no cloud required
• Local-first - Your memory stays under your control
• Offline-capable - Works without internet (except bridge mode)
• PWA - Install on your phone like a native app
• Open source - Full transparency and control