Why we built Dayglass: 100% local, bot-free Mac meeting & screen MCP

Hey Product Hunt! Ahead of our Dayglass Hypership launch tomorrow (Nov 6 at 12:01 AM PT), wanted to share the architecture behind why we built a 100% local, bot-free meeting & screen memory engine for macOS:

1. Zero creepy meeting bots: Instead of inviting an Otter/Fireflies/Fathom bot into your Zoom, Teams, or Meet calls, Dayglass captures system audio + mic + Apple Vision screen OCR natively on Apple Silicon.

2. Solving the 6–26 GB/user KV-cache context bottleneck: Revolut just revealed they cut AI coding costs 90% by making frontier Claude the exception, but hit a 6–26 GB/user GPU RAM bottleneck just for context retention. Dayglass stores a 3-layer Unified Context Graph (short_term, long_term, decision_trace) in <225 MB of local SQLite and exposes 26 MCP tools to Claude Desktop, Cursor, and Codex in <15ms.

3. Action-Risk Governance: Paired with ThumbGate (https://thumb-gate.app), every thumbs-down or correction compiles into a deterministic PreToolUse guardrail and 6-Block EARS SDD spec.

Try Dayglass free (`brew install --cask iganapolsky/tap/dayglass`) or grab Dayglass Pro ($19/mo or $79 lifetime at https://dayglass.io/pricing). What MCP workflows are you running with local screen & call memory?