Dayglass is a macOS application that acts as a private AI memory for your work day. It records and transcribes your Zoom, Microsoft Teams and Google Meet calls directly on your Mac without a bot ever joining the meeting, and it keeps the resulting transcripts, screen text and notes in a local database under ~/.dayglass. The product's stated purpose is to give Claude Code and Cursor context from your meetings while the call is still happening, so decisions can be turned into an implementation plan. It is built for engineers and founders who hold sensitive engineering and customer conversations and do not want a cloud meeting bot sitting in the room. Dayglass runs on macOS 13 or later on Apple Silicon, can be installed from a free DMG or through Homebrew, and is free for your first ten meetings.
The problem Dayglass addresses is stated directly on its website: stop re-explaining client calls to Claude Code. Meeting decisions and commitments are made verbally, and by the time an engineer wants a coding agent to act on them, that information has to be reconstructed from memory, from a follow-up email, or by digging through notes. Dayglass describes this as follow-up email archaeology, asking what was agreed instead of digging for it. At the same time, the founder built Dayglass after getting tired of cloud meeting bots joining sensitive engineering and customer calls. Cloud notetakers process and store recordings and notes on vendor servers, and many join the call as a bot by default. Dayglass positions itself against that model by keeping capture and storage local and by letting you choose what, if anything, leaves the machine.
The core capability is a live meeting feed for coding agents. When a meeting is in progress, Claude Desktop, Claude Code, Cursor and the other agents on your Mac can call the local MCP tool dayglass_get_live_meeting and read who is talking and what was just said. The MCP server runs on your Mac, and an installation command, dayglass doctor --install-mcp, is provided. The website illustrates this with a terminal session in which the user asks Claude what Alex just committed to on the call, and Claude answers that Alex committed to having the retry change ready by Thursday 4 PM, that Mina asked for it, and that Sam will confirm the design review. The agent reads the speaker-tagged transcript, the people on the call and the shared screen while the meeting continues.
Recording is bot-free. Dayglass captures both sides of Zoom, Google Meet and Teams calls directly from macOS system audio, including AirPods, and from your microphone. It uses native ScreenCaptureKit to record on Apple Silicon, with audio captured as 16 kHz PCM WAV and transcribed on the Mac with local Whisper. No recording bot joins the call, so nothing appears in your calendar invites. When a call ends, Dayglass extracts key decisions, action items by owner and a ready-to-send follow-up email draft, generating the note locally two seconds after hanging up. Notes can be exported as Markdown with one click. Search covers every call and everything shared in it, including shared slides, terminal stack traces, documents and spoken quotes, captured through meeting screen and slide capture plus Apple Vision OCR and SQLite FTS5.
Additional capabilities handle the human side of meetings. Live Call Catch-Me-Up, triggered with Command-Shift-Space, shows what was just proposed, how it connects to prior meetings and the sharpest questions to ask next, which helps when you join late or zone out. Pressing Command-Control-B while debugging or shipping summarizes the exact steps you took into a clean checklist or reusable agent skill, turning a 15-minute focus block into a reusable playbook saved as a Starred Work Session. A built-in one-click call consent snippet copies a friendly one-liner to drop into Zoom or Meet chat so participants know you are taking local notes. Ask AI can run on a local model or a cloud model, whichever you pick.
The overall approach follows a four-step local pipeline described on the site. macOS capture uses ScreenCaptureKit for audio and Apple Vision neural OCR for screen text. Local SQLite memory stores everything with SQLite FTS5 and hybrid search inside ~/.dayglass/dayglass.sqlite. Privacy guards skip windows of password managers such as 1Password and Bitwarden during capture and withhold Never-share terms, by default, before any text goes to a cloud model you choose. Your workflow then receives clean Markdown notes plus the Claude and Cursor MCP connection. Because every table lives in plain SQLite, the database can be inspected with sqlite3, and network traffic can be checked with Little Snitch or lsof -i. An optional continuous screen memory setting captures your screen outside meetings too, so you can search what you saw last week, and it can be switched on or off in settings.
The outcomes the site claims are practical. Every call ends with who decided what, which is the stated benefit of decisions that don't get lost. Instead of follow-up email archaeology, you can simply ask what was agreed. Your agent already knows what was decided, because Claude and Cursor read the call as it happens. For people who discuss sensitive engineering and customer work, Dayglass turns privacy into a selling point: transcripts, screen text and notes stay on the Mac, audio is transcribed on the Mac and not kept by default, and the only thing that can leave is text sent to a cloud model, and only if you turn one on. With a local model, nothing leaves the Mac. The site also states that Pro and Lifetime purchases carry a 14-day money-back guarantee handled by email.
The clearest documented use case is a client call that changes a requirement, for example retrying the payment webhook two times instead of five. While the call is still running, you ask Claude Code or Cursor to plan that change, it reads the live transcript from Dayglass over MCP, and it works from what was said so you do not retype what the client asked for. The site labels this an illustration with made-up details and notes that results depend on your call, your agent and your code. Other documented workflows include catching up on a call you joined late, searching back through a shared slide, terminal stack trace or spoken quote after the meeting, drafting the follow-up email from the extracted action items, and summarizing a debugging session into a reusable checklist or agent skill.
Dayglass is aimed at engineers and founders. It was built by AI systems engineer Igor Ganapolsky, whose Apple Developer ID is 9GMM26JC5X, after getting tired of cloud meeting bots joining sensitive engineering and customer calls. Integrations named on the site are Zoom, Google Meet, Microsoft Teams, Claude Desktop, Claude Code and Cursor. The stack described includes macOS ScreenCaptureKit, local Whisper, Apple Vision OCR, SQLite FTS5, a JSON-RPC MCP server and the Dayglass DB. Pricing starts with a free tier covering your first 10 meetings with no card, including Ask AI, AI briefs and the Claude/Cursor live link; after that, capture, transcripts, search and export stay free while Ask AI on older history, AI briefs and the Claude/Cursor live link need Pro. Pro is $149 a year or $19 a month, with a founding Lifetime price of $199 once for the first 100 buyers, then $249.
Dayglass makes a specific promise: private, bot-free meeting notes on your Mac that your coding agent can read while the call is live. It combines local capture through ScreenCaptureKit, on-device transcription, an inspectable SQLite memory with searchable screen text, automatic decisions and action items, and a local MCP server that feeds Claude Code and Cursor. For engineers and founders who treat meeting detail as sensitive, it keeps the record on the machine while still making the conversation usable by the tools they already work in.