A finished research run on an example topic, shown in full — the findings, the sources behind them and the product versions are all the run's own output, except that a product name may have been changed where it matched an existing brand.

Back to the run overview AI meeting notes

Experimental

Consensus Protocol

An open local data standard and lightweight sync engine that lets any local-first notetaker (Meetily, Hyprnote, or the many solo repos) export to a shared format, enabling interoperability across the fragmented open-source ecosystem.

01

What to build

  • MVP

    A published open schema plus a reference CLI/library that converts output from three or four popular local notetaker repos into one common JSON transcript+diarization format, with a simple local search index built on top.

  • Why it can be better

    The open-source scene is a long tail of single-maintainer, single-OS repos with no shared format, so users accumulate disconnected local histories; a shared standard turns fragmentation into an interoperable ecosystem rather than requiring one company to rebuild every capture mode itself.

  • Revenue model

    Open-core model: the standard and CLI are free, with a paid hosted (still local-encrypted) sync and search app for teams wanting cross-device consolidation.

  • Target user

    Developers and privacy-focused power users already running open-source local notetakers who want their tools to interoperate.

02

Context

The opportunity this answers

Build a local-first meeting-memory layer for professionals that unifies transcripts from any notetaker or capture mode into one private, searchable history.

Why now

A wave of open-source local-first notetakers built on Rust/Swift/Tauri has already proven demand for on-device processing, with diarization and Whisper models now standard, high-download building blocks anyone can bundle, while search interest for generic 'AI meeting notes' has plateaued — signaling the crowded cloud tier is saturated even as privacy-driven local alternatives keep multiplying, and none of them yet unify capture contexts, ingest rival platforms' exports, or sync back out privately.

Trend & recent change

7 findings19 sources

Back to the idea

AI meeting notes is shifting from cloud SaaS notetaker bots toward local-first, privacy-preserving alternatives, with a wave of open-source repositories built on Rust/Swift/Tauri stacks offering on-device transcription and diarization; incumbent platforms (Google Meet, Notion, Otter, Read AI) are simultaneously embedding notetaking natively, while search demand for the category has softened slightly over the past several months.

  1. 01

    Local-first note-taker wave

    A cluster of new open-source, local-first meeting note apps has emerged, explicitly marketed against cloud incumbents on privacy grounds

    Evidence

    Meetily (31k stars) bills itself as '100% local processing, no cloud required' built on Rust with Parakeet/Whisper transcription and Ollama summarization; Natively positions itself as 'the best alternative to Cluely, Otter, Granola, Final Round AI, Fireflies'; humla, os-clovy, note67, and phorminx all advertise on-device transcription, offline speaker ID, and no-subscription models.

    Sources github.com (opens in a new tab) github.com (opens in a new tab) github.com (opens in a new tab) github.com (opens in a new tab) github.com (opens in a new tab) github.com (opens in a new tab)

  2. 02

    Diarization goes mainstream

    Speaker diarization models have become a standard, high-traction component rather than a niche add-on

    Evidence

    pyannote/speaker-diarization-3.1 has 7.4M downloads and 3,895 likes, and a newer pyannote/speaker-diarization-community-1 already has 5.5M downloads and 2,109 likes, indicating diarization is now a core building block bundled into meeting-note pipelines alongside Whisper-family ASR models (openai/whisper-large-v3-turbo, whisper-large-v3) that also show multi-million download counts.

    Sources huggingface.co (opens in a new tab) huggingface.co (opens in a new tab) huggingface.co (opens in a new tab) huggingface.co (opens in a new tab)

  3. 03

    Platform-native notetaking arrives

    Major productivity and video-conferencing platforms are folding AI notetaking directly into their core products, reducing reliance on third-party bots

    Evidence

    Google Meet now lets admins turn on Gemini-powered AI note-taking that auto-generates a Google Doc attached to the calendar event; Notion ships 'AI Meeting Notes' that routes recaps to Slack/email and assigns action items via its own agent.

    Sources knowledge.workspace.google.com (opens in a new tab) notion.com (opens in a new tab)

  4. 04

    Bot-free, in-person capture expansion

    Tools are expanding beyond calendar-linked video meetings into in-person, bot-free, and cross-context capture

    Evidence

    Krisp promotes a 'bot-free AI Note Taker' for both online and in-person meetings; Otter's notetaker can 'record bot-free on desktop or Chrome' or capture on mobile; Summary AI and Minutes market support for lectures, phone calls, podcasts, and uploaded audio/video in addition to meetings.

    Sources krisp.ai (opens in a new tab) otter.ai (opens in a new tab) pub.towardsai.net (opens in a new tab) apps.apple.com (opens in a new tab)

  5. 05

    Query-your-meetings features spread

    Vendors are converging on 'answerable transcript' features — asking natural-language questions over meeting history — as a differentiator

    Evidence

    Read AI enables 'Ask Read' to find answers across meetings, emails, and messages; humla lets users 'ask your notes and get cited answers'; Minutes lets users ask questions like 'What stats did Laura reference on her slide?'

    Sources read.ai (opens in a new tab) github.com (opens in a new tab) apps.apple.com (opens in a new tab)

  6. 06

    Search demand plateauing

    Search interest for the category has been flat-to-declining across recent months rather than growing

    Evidence

    Average monthly volume sits near 1,000 with a recent trajectory of 1600 (Mar), 1000 (Apr), 1000 (May), 880 (Jun), 1000 (Jul), 880 (Aug), showing a pullback from an earlier spike rather than sustained growth.

  7. 07

    Crowded competitive benchmarking content

    Comparison and review content treating the category as crowded and mature is proliferating, with new entrants explicitly benchmarking against a fixed set of incumbents

    Evidence

    A hands-on review compares 8 tools including Otter, Fathom, and Fireflies plus an 'agent that acts on your notes'; a Reddit thread comparing Fireflies against Sembly AI shows practitioners actively cross-shopping multiple named tools.

    Sources simular.ai (opens in a new tab) reddit.com (opens in a new tab)