A workspace for building apps against Everysight's Maverick AI smart-glasses SDK
(alpha), developed against the desktop Glasses Simulator until physical hardware
ships. Not a single app - each idea gets its own independent project under
projects/.
If you're an AI coding agent, read CLAUDE.md first (or AGENTS.md, which just
points to it) - it has hard rules and a reading order, kept short on purpose.
CLAUDE.md/AGENTS.md- entry point for coding agents: hard rules and reading order. Humans can skim it too, but it's written for an agent picking this workspace up cold.DEV_NOTES.md- hands-on gotchas that apply across any project here (toolchain setup, real SDK API corrections, simulator behavior, Wear OS quirks, target hardware). Read this before starting a new project, and add to it when something new comes up that isn't specific to just one app.MAVERICK_SDK_NOTES.md- reference crawl of Everysight's official SDK documentation site (architecture, API surface, UI Kit, simulator setup, etc).VISION.md- conceptual architecture and feature roadmap for the personal, modular glasses OS this workspace is building toward - read before starting new feature work.EDGE_AI_ARCHITECTURE.md- technical companion toVISION.md's AI layer: fine-tuning strategy, local knowledge base (SQLite/vector hybrid), vision pipeline, context management, and voice stack (TTS/VAD engine choices with RAM/latency specs for the S26 Ultra).simulator/- the Glasses Simulator binary + supporting files (gitignored, large binary - re-download if missing, see DEV_NOTES.md).projects/- each subdirectory is a fully independent Gradle project (own build files, ownNOTES.mdfor anything specific to that app, ownCLAUDE.mdpointing back up to the workspace-level one):heart-rate-monitor/- phone app showing live heart rate from a Wear OS watch on the glasses HUD.song-id/- music recognition triggered entirely from the glasses (tap to listen via the glasses' own mic, swipe to browse history), provider-agnostic recognition backend (AudD first). Replaces an earliershazam-glassesprototype that tried to hijack the real Shazam app - seesong-id/NOTES.mdfor why that didn't work.voice-agent/- standalone (no Maverick SDK yet) on-device voice-conversation spike, branded "TryptaOS": VAD barge-in, Gemma 4's native audio input (no STT in the response path), tunable local TTS, live waveforms, session logging. Forked fromgoogle-ai-edge/galleryrather than built from scratch - seevoice-agent/NOTES.mdfor the real, tested state and exact working config before assuming anything here is still speculative. This is whereEDGE_AI_ARCHITECTURE.md's "voice stack" section became a real implementation instead of a research plan.
- Read
CLAUDE.mdandDEV_NOTES.mdfirst. - Create
projects/<name>/with its own Gradle setup (seeheart-rate-monitor/gradle/libs.versions.tomlfor known-working dependency versions to start from). - Give it its own
NOTES.mdfor anything project-specific, and a shortCLAUDE.mdpointing back to the workspace root one (copyheart-rate-monitor/CLAUDE.mdas a template - it's the same three lines for every project). - If you discover something that isn't specific to that one app, add it to the root
DEV_NOTES.mdinstead, so the next project benefits too.