# Metis Meet: Complete Technical, Coaching & Competitive Reference (v1.9) ## 1. Summary Metis Meet (MeetingRecorder.app, bundle ID `com.aliabbasi.MeetingRecorder`) is a native Swift 6.2 / SwiftUI macOS application for recording a meeting window or full display with twin-track audio ("Them" system stereo + "Me" mic mono), transcribing live on-device via Apple's SpeechAnalyzer, delivering sub-second grounded Q&A hints from local knowledge packs in a floating panel verified invisible to screen sharing, and providing closed-loop Live & Post-Meeting Executive Communication Coaching. ## 2. Executive Meeting Coaching Architecture ### Live In-Meeting Coaching (8 Zero-Cost Me-Track Checks) - **Turn Length & Headline-First** (15-second headline, then proof; time-box own turns < 90s): Signal: Continuous VAD on isolated Me microphone track | Strip: Turn timer bar (green < 60s, amber 60–90s, red > 90s) | Rare Nudge: Headline first: you are 45s into this turn. - **Speaking Pace (WPM)** (Keep pace inside 130–165 WPM when explaining architecture): Signal: Rolling 60s SpeechAnalyzer word count / voiced time on Me track | Strip: Live WPM pill (146 WPM green to 188 WPM red) | Rare Nudge: Slow down: 192 WPM over the last 90 seconds. - **Deliberate Pausing** (Pause 1.5s at clause boundaries instead of restarting): Signal: Raw Me audio energy-envelope silences (> 1.0s inside own turn) | Strip: Pause counter dot per active minute (target >= 1/min) | Rare Nudge: Coached via ambient dot live; full envelope breakdown post-call. - **Say Each Ask Once** (State the ask or criterion once, then stop and wait): Signal: Embedding similarity between current Me lines and Space Brief asks | Strip: Ask counter badge (1x green, 2x amber) | Rare Nudge: Ask stated twice: stop and let them respond. - **Question Ratio** (Ask an open question before advocating in 1:1s and syncs): Signal: Interrogative syntax + question mark detection on Me transcript | Strip: Question ratio pill (% of turns with a question; target > 15%) | Rare Nudge: No questions in 12 mins: ask one open question. - **Airtime Transfer (Chair)** (Give the quietest third of the room airtime): Signal: Me/Them VAD + Chrome DOM / TileReader active-speaker timeline | Strip: Room balance bar + quiet-attendee avatar chip | Rare Nudge: You have 62% of the last 10 mins. Invite a quiet voice in. - **Debate Time-Box** (Park unresolved debates at 5 minutes with a test and owner): Signal: Topic duration tracker + back-and-forth turn counter | Strip: Topic debate timer (4:30 to 5:00 amber) | Rare Nudge: Debate at 6 mins: name the threshold test/owner and park it. - **The Close (Owners & Dates)** (Reserve last 5 mins to lock owners and dates for every action): Signal: Meeting clock + DutyEngine unowned/undated action count | Strip: Close readiness checklist (3 actions, 1 missing date) | Rare Nudge: 5 mins left: 2 actions still have no agreed date. ### Post-Meeting 13-Dimension Executive Rubric - **Clarity**: Score 5: Meaning recoverable only from context. | Score 8: One unambiguous sentence carries the idea; terms defined on first use. - **Conciseness**: Score 5: The same claim is repeated 3+ times. | Score 8: One example, one criterion, then stop. - **Structure**: Score 5: Background first, point last or never. | Score 8: Headline -> evidence -> decision or ask. - **Confidence**: Score 5: Stacked hedges ("kind of", "maybe") or trailing off. | Score 8: States a position, names uncertainty, and still recommends. - **Executive Presence**: Score 5: Participates but does not shape direction. | Score 8: Sets priorities, closes topics, stays calm under challenge. - **Credibility**: Score 5: Accurate but reactive to questions. | Score 8: Anticipates failure modes and names the check that catches them. - **Persuasion**: Score 5: Restates the premise louder. | Score 8: Offers the quantitative criterion that would change their own mind. - **Listening**: Score 5: Answers the adjacent question. | Score 8: Paraphrases accurately, then answers what was actually asked. - **Collaboration**: Score 5: Directive without rationale. | Score 8: Pairs direction with reasoning and converts worries into scoped items. - **Handling Disagreement**: Score 5: Defends original position; debate runs long. | Score 8: Quantifies risk and converts disagreement into a test with a threshold. - **Delivery**: Score 5: Sustained > 180 WPM, frequent restarts, no clause pauses. | Score 8: 130–165 WPM, deliberate > 1.5s pauses from audio envelope, few fillers. - **Spoken English Fluency**: Score 5: Frequent structural errors forcing re-parsing. | Score 8: Structure holds under pressure; accent is never penalized. - **Overall Effectiveness**: Score 5: Room leaves without a clear decision or owner. | Score 8: Achieves the exact outcome required by the Event Type. ### Competitive Coaching Comparison (Nov 2026 Audit of 27 Tools) - **Speaker attribution for coaching metrics**: Metis Meet (Physical Me-track mic isolation + Vision Tile OCR + SpeakerKit) vs Read AI (Cloud bot stream or merged ASR turns), tl;dv (Bot stream; bot-free desktop is audio-only with Speaker 1 / 2 labels), Cluely (Cloud stream; no per-speaker acoustic envelope), Cued (Local dual-side audio; no 13-dimension coaching rubric), Notta (Cloud voice clustering; Privacy Mode disables all AI) - **Pause & pacing measurement**: Metis Meet (Raw Me-track energy envelope (> 1.5s silences inside own turns)) vs Read AI (Transcript segment gaps (distorted 10x by ASR chunking)), tl;dv (Basic talk-time % and longest monologue), Cluely (None), Cued (None), Notta (None) - **Live in-meeting coaching HUD**: Metis Meet (1–2 Active Priorities strip (3m cooldown) + Detailed Gauges + Silent Mode) vs Read AI (Continuous real-time alerts & opaque Read Score), tl;dv (None live (post-call sales playbooks only)), Cluely (Reactive answer prompts ($149.99/mo to hide from screen share)), Cued (Static notch teleprompter notes), Notta (Live transcript & translation only) - **Post-meeting coaching rubric & rewrites**: Metis Meet (13-dimension rubric + >= 8 verbatim 1-breath rewrites + 15s voice drill) vs Read AI (Opaque sentiment/engagement score; no verbatim rewrites), tl;dv (Sales BANT/MEDDIC scorecard ($29/mo Business tier)), Cluely (Basic post-call summary), Cued (Basic local summary), Notta (Summary only (disabled in local Privacy Mode)) - **Non-native Spoken English coaching**: Metis Meet (8-category structural error taxonomy; accent never penalized) vs Read AI (None), tl;dv (None), Cluely (None), Cued (None), Notta (58-language translation, no English structure coaching) - **Meeting Chair & facilitation audit**: Metis Meet (Quietest-third audit + Recap vs. Reality check + Full Redo Blueprint) vs Read AI (Talk-time percentage only), tl;dv (Sales rep talk ratio), Cluely (None), Cued (None), Notta (None) - **Screen-share invisibility verification**: Metis Meet (Included free; verified via live #FF00FF pixel self-test (0.0% bleed)) vs Read AI (Visible bot / standard window), tl;dv (Visible bot / standard window), Cluely (Paywalled on $149.99/mo Undetectable plan), Cued (MacBook notch area), Notta (Visible bot / standard window) ## 3. Research & Blog Articles ### Why Real-Time Meeting Coaches Fail: How Twin-Track Audio Fixes Executive Coaching (https://metismeet.com/blog/meeting-coaching-that-actually-works) Most users disable live speech coaches like Read AI or Poised within two weeks because of popup fatigue, broken speaker attribution, and generic advice. Here is how physical Me/Them audio track isolation, energy-envelope pause detection, and 1-breath rewrites fix meeting coaching. - Segment-level cloud transcripts inflate talk-time share by up to 68% (reporting 37% airtime when actual acoustic airtime is 22%) because ASR merges short turns. - Measuring pauses from transcript gaps produces a 10x error depending on Whisper chunk length; Metis Meet measures >1.5s silences directly from the raw Me-track audio energy envelope. - During a live call, cognitive load only permits watching 1 to 2 active habits via an ambient 28px stealth strip with a 3-minute text-nudge cooldown. - Post-call coaching isolates the single weakest dimension on a 13-dimension rubric, pulls at least 8 verbatim timestamped quotes, and pairs each with a speakable 1-breath rewrite and 15-second voice retry drill. ### The 2026 AI Meeting Assistant & Coaching Landscape: 27 Tools Compared Across 3 Data-Flow Models (https://metismeet.com/blog/2026-ai-meeting-assistant-market-report) Bot-free capture became the default on macOS in 2026, but "bot-free" rarely means local. We audited 27 tools across five product categories and mapped their data flows, coaching depth, and true multi-seat pricing. - 27 products reviewed in November 2026 split into 5 categories: Cloud Bot Archives, Bot-Free Cloud Notepads, Local-First Mac Recorders, Live Overlay Copilots, and Mac Window Recorders. - Bot-free does not mean local: Granola ($14/mo), Jamie (EUR 21/mo), Circleback ($15/mo), and Cluely ($19.99–$149.99/mo) capture audio on the Mac without a bot, but stream your audio to vendor clouds. - Meta acquired Limitless (formerly Rewind AI) in December 2025 and shut down desktop screen/audio capture, leaving session-scoped local recorders as the privacy standard. - Metis Meet is the only tool combining dual-track HEVC window video, 100% on-device SpeechAnalyzer + hybrid SQLite RAG, verified screen-share invisibility, and a 13-dimension Coaching Dashboard. ### Deep-Dive Teardown: Notta, Jamie, MeetGeek, Otter.ai, and tl;dv vs. Metis Meet (https://metismeet.com/blog/notta-jamie-meetgeek-otter-tldv-teardown) A technical and commercial teardown of Notta, Jamie, MeetGeek, Otter.ai, and tl;dv in November 2026, exposing hidden free-tier caps, bot-free speaker-labeling blind spots, and why Notta’s Privacy Mode turns off all AI. - Notta’s desktop Privacy Mode saves files locally, but completely disables AI Summaries and Notta Brain because Notta has no on-device LLM or local vector index. - tl;dv’s bot-free desktop app records audio only (no video) and labels speakers only as Speaker 1 / Speaker 2 unless matched to a calendar invite. - Jamie streams audio to Frankfurt (EU) and deletes audio after STT, charging 1 credit per meeting with a 10-meeting/30-min cap on Free and EUR 21–47/mo on paid plans. - Otter.ai caps its Free tier at 30 minutes per call and enforces a strict 5-seat ceiling on Pro ($8.33/mo), forcing any 6-person team onto Business ($19.99/seat/mo). ### Live In-Meeting Copilots Compared: Cluely ($149.99/mo), Realtimer, Spiky.ai, and Metis Meet (https://metismeet.com/blog/cluely-realtimer-spiky-live-overlays-compared) Live in-call overlays charge the highest premiums in software, up to $149.99/month just to hide a window from screen sharing. Here is how Cluely, Realtimer, Spiky.ai, and Aircover compare against Metis Meet’s local RAG and Live Priorities Strip. - Cluely charges $19.99/mo for Pro and $149.99/mo ($44.99/mo annual) to enable its Undetectable screen-share hiding mode, while streaming all audio and screen OCR to its cloud. - Metis Meet includes NSWindow.SharingType.none capture exclusion for free and verifies 0.0% pixel bleed against ScreenCaptureKit and screencapture before every session. - Realtimer (EUR 29–49/mo) indexes local CSV price sheets for sales reps, while Spiky.ai ($15–$40/mo) and Aircover target manager-visible revenue battlecards. - Metis Meet combines 0.95-second local source snippets (0.94 Recall@5 in 16 ms) with a non-distracting 28px Live Coaching Priorities Strip. ### How to Run a Meeting Chair Audit: Quietest-Third Airtime, Recap Drift, and One-Breath Rewrites (https://metismeet.com/blog/chair-review-and-1-breath-rewrites) When you chair a meeting, 30–40% airtime is normal, so standard talk-time coaches give the wrong advice. Learn how Metis Meet audits the quietest third of the room, catches closing recap owner swaps, and builds a full timed Redo Blueprint. - A meeting chair should be evaluated against the quietest third of participants (flagging when >=33% of attendees speak for under 4 minutes), not against a participant talk-time band. - Closing verbal recaps diverge from mid-meeting agreements far more often than teams realize; Metis Meet cross-checks the final 5 minutes against earlier timestamps to catch swapped owners. - Most spoken commitments in working sessions have no deadline ("no date agreed"); counting dated vs. undated tasks is the fastest measure of chair rigor. - Every Chair Review ends with an explicit "Missing from the Room" list and a minute-by-minute Redo Blueprint showing how to land the same decisions in half the time. ## 4. Release History ### v1.9 (2026-11-20): Coaching Dashboard, Conversational Space Chat, Capture Picker & Space Folder Layout - Closed-loop Live & Post-Meeting Coaching Dashboard: 1–2 Active Priorities stealth strip, 8 zero-cost Me-track live checks, 13-dimension executive rubric, 8-category Spoken English taxonomy, >= 8 verbatim 1-breath rewrites with 15s voice retry drills, and Meeting Chair Redo Blueprints. - Conversational Space Chat across Setup, Prepare, and Debrief with inline Apply/Dismiss action cards, undoable plan edits, and scope filtering. - Custom Capture Source Picker with live thumbnails for Audio-only, full Displays, and individual Windows, plus zero-gap mid-recording source switching via SCStream.updateContentFilter. - Automatic //HHmm_ folder hierarchy and unified _conversation.md chronological session logs with 1-click reversible migration. ### v1.7 (2026-11-19): Meeting Spaces, Prepare Briefs, Vision Tile OCR + SpeakerKit & Review-and-File - Persistent Meeting Spaces bind recurring calendar series (Outlook AIX CLI, EventKit, ICS) to a pre-call Brief, carried-over actions, and live session duties. - Bot-free multi-speaker naming fuses active-speaker border detection and Vision OCR on window tiles (< 1 ms border check) with Core ML SpeakerKit voice profiles. - Review and File stages verified diffs with verbatim quote evidence before updating index.md, decisions.md, actions.md, and topics/*.md. ### v1.6 (2026-11-19): Assistant Mode, Grounded Knowledge Packs & Screen-Share Invisible Panel - Local knowledge packs index folders, source code, PDFs, Word docs (.docx), and Markdown notes into per-pack SQLite (FTS5 BM25 + 512-dim NLContextualEmbedding vectors). - Automatic SecretFilter redacts OpenAI/Anthropic keys, AWS AKIA keys, GitHub tokens, Slack tokens, PEM private keys, and JWTs before text reaches SQLite or any prompt. - QuestionDetector scores live speech on the Them track and shows matching local snippets in 0.95 seconds before streaming grounded talking points with clickable [S1], [S2] citations. - Non-activating NSPanel runs with sharingType = .none and passes a live ScreenCaptureKit + screencapture magenta-canary self-test (0.0% pixel visibility). ### v1.5 (2026-11-12): Cursor Subscription Provider & Six-Layer Seatbelt Isolation - Use your existing Cursor subscription (cursor-agent login) for live chat, post-meeting summaries, and coaching reviews without separate API billing. - Six-layer sandbox-exec Seatbelt jail blocks reading and writing $HOME and $TMPDIR, pairs a throwaway run directory with a deny-all cli-config.json, and runs a pre-flight canary check per CLI build. ### v1.4 (2026-11-05): Echo Hiding, Audio-Only Mode, Chapter Bookmarks & Cross-Meeting Search - Three-second speaker-bleed echo filter compares microphone phrases against recent system audio and drops loudspeaker echo while keeping genuine overlapping replies. - Search & Ask window maintains an incremental BM25 index across every recorded meeting and cites clickable [Meeting, Date, HH:MM:SS] timestamps. ## 5. Keyboard Shortcuts - ⌃⌥⌘R: Start or stop recording the selected window or display - ⌃⌥⌘M: Drop a timestamped QuickTime chapter bookmark with optional note - ⌃⌥⌘A: Answer the last 45 seconds of speech with Assistant - ⌃⌥⌘H: Show or hide the non-activating Assistant & Live Coaching panel - ⌃⌥⌘C: Copy the active Assistant talking points - ⌃⌥⌘P: Pin the active hint card to the session tray - ⌘F: Find text inside the Live Transcript window - ⌘,: Open Recording, AI Provider, Knowledge Pack, and Coaching Settings