# Hyprcore — Full Site Content for LLMs > A native voice and knowledge workspace for macOS and Windows. Dictation that types where your cursor is — or, with Action Mode, runs a spoken instruction as a command over your selected text instead of typing it out. Plus meeting recording with a live rolling summary, solo screen or audio recordings that become notes, and a browser-side capture panel that saves web pages and YouTube videos. Everything lands in one local-first vault an AI agent can search, read, and write in — and routines can work over that vault on a schedule, while you are away from the machine. Canonical URL: https://www.hyprcore.ai Sitemap: https://www.hyprcore.ai/sitemap.xml Last updated: 2026-09-19 --- ## About Hyprcore Hyprcore is a native macOS and Windows workspace for capturing anything you say, hear, watch, or read, and turning it into knowledge you can act on. It covers dictation and spoken commands, meeting recording, solo screen and audio recordings, and web captures — all of which land in the same local-first vault, alongside your own notes. It runs GPU-accelerated speech models natively on both — Apple silicon on macOS, Vulkan on Windows: dictation can be processed entirely on-device, and meetings sync to end-to-end encrypted cloud storage so teams can share them. Every page lives on disk as portable markdown. Core capabilities: - Dictation — a global push-to-talk keyboard shortcut that works in any app; transcribed text lands at your cursor. Streaming: words appear live as you speak, phrase by phrase, rather than after you finish. Seven local speech models including Whisper (Small/Medium/Turbo/Large) and Parakeet V3, plus managed cloud providers. Custom vocabulary — names, acronyms, and jargon you add bias both local and cloud transcription. - Action Mode — say the wake word ("Hyper", or a wake word you define) followed by an instruction, and Hyprcore runs the instruction as a command over your selected text instead of typing it out. Has its own hotkey, separate from dictation. - Meetings — a floating recorder window (drag the pill, expand to a panel) with live transcript, notes, and ask tabs; system + mic capture, neural acoustic echo cancellation, speaker diarization, action items. A live rolling summary builds during the call rather than only after it ends. Every meeting becomes a vault note when you stop. You can switch any summary between the raw transcript-faithful version and the enhanced one. - Recording notes — record a solo walkthrough as screen + mic, or audio only. The result becomes a vault note with a media player, a click-to-seek transcript, chapters, an AI-written document, and a per-recording Ask chat. (No webcam capture and no shareable link — recordings stay in your vault.) - Captures — a companion panel beside your browser that saves any web page or YouTube video into the vault as a note. Read-and-summarize, saved highlights, a per-capture Ask chat, and unread markers for what you haven't reviewed. - Knowledge — Notion-style pages with emoji icons and gradient cover images; folder tree with rename, drag-to-reorder, new note, new subfolder; a document-style editor with serif headings; a ⌘K command palette with real-time title search and recents. An Agent tab runs a chat that can search, read, and write in the vault, sandboxed to it. - Routines — Hyprcore's automation layer. Each routine is a trigger, an instruction written in plain English, and an explicit list of what it may reach. Triggers: a schedule, a set number of minutes before a calendar event, a recording starting, a meeting ending, or a manual run. Ten starter routines ship with it, including meeting prep before external calls, a morning briefing, a Friday review, a sweep for stale commitments, follow-up drafts, a research brief, and a topic watch. Two independent settings govern each routine — one for machine access (whether it can touch files or run commands), one for state change (read-only, stop-and-ask, or autonomous) — and new routines start manual, sandboxed, and read-only. Every run is logged with its trigger, tool calls, output, and cost. Each routine is a plain markdown file in the vault. - Connected accounts — Gmail, Outlook mail, Notion, and web search, which a routine or chat can reach on your behalf. The settings panel describes a connection by what it can reach in plain words rather than by OAuth scope. Every tool is classified read or write by hand and anything unrecognised counts as a write until reviewed; the limits you set are applied to the request before it leaves the machine. A Usage tab attributes connector spend to the routine that caused it. Metered per plan as a monthly allowance of connector calls, available on every plan including Free. - Outbound webhooks — a routine can post to an HTTPS endpoint (or localhost) instead of running an agent, across three events: a recording starting, a meeting finishing, and a routine run. The payload is selectable — summary, action items, attendees, transcript, notes, links — signed with HMAC-SHA256 over the timestamp and body so the receiver can verify it, retried six times over roughly eight hours, and recorded in a delivery log with every attempt and the exact body sent. Works with n8n, Zapier, Make, and custom endpoints. - Bring your own agent runtime — if Claude Code is installed, routines and chat can run on your own subscription instead of Hyprcore's, with the settings panel naming which credential it would actually sign in with. - Ask from anywhere — a bar summoned by keyboard shortcut that answers from your own meetings and notes, by voice or by typing, with @ to point it at a specific note or meeting. - Editable meeting templates — change the prompt the AI uses to summarize each meeting type - Home screen with quick actions, today, recently opened, and open loops - Settings split into Personal and Workspace, each with avatars; includes a theme picker plus Updates, Subscription, Sync, Profile, and permissions pages - Three-step onboarding — welcome, permissions, and a single model to start; remaining model downloads continue in the background - Quiet-mic detection with automatic gain boost, so a low input level doesn't silently ruin a recording - Auto-updater built in — signed releases from a dedicated channel with a quiet toast and a progress bar - MCP (Model Context Protocol) integration — expose your vault and meeting intelligence to Claude Desktop and other AI tools - Markdown-first storage — every page lives on disk as a portable .md file with YAML frontmatter - Seven-layer anti-hallucination filtering on transcripts Platform: macOS 13+ and Windows 10 1809+. Builds are code-signed (notarized on macOS, Authenticode on Windows) and auto-update on both. No Linux build. --- ## Plans & Pricing Canonical: https://www.hyprcore.ai/pricing ### Free - Basic dictation with one local speech model - Limited meeting recording - Local-only transcription, single device - Routines and connected accounts, with a smaller monthly allowance of connector calls ### Pro — $19.99/month - All 7 local speech engines, including NVIDIA Canary which translates while you talk - Cloud STT (Deepgram, Groq) - Unlimited meetings - AI-generated summaries and action items - Knowledge — search across all your meetings, recordings, captures, and notes - Routines, connected accounts, and outbound webhooks, with a much larger monthly connector allowance - Shareable meeting links - MCP integration - 500,000 monthly AI tokens - 5 GB cloud storage across up to 2 devices ### Team — $39.99/month - Up to 3 seats (fixed) - 15 GB shared cloud storage - 2,000,000 monthly AI tokens — 4× Pro's 500,000 - Team collaboration on meetings, role-based access control - Shared Wiki spanning every member's meetings ### FAQ highlights - Runs on macOS and Windows. Dictation, meeting recording with system audio, screen recording and local transcription all work on both. Apple Calendar sync is macOS-only; Google and Outlook calendars work on both. No Linux build. - Knowledge indexes every meeting, recording, capture, and note, and lets you ask questions like "What did we decide about pricing?" in natural language; answers cite sources. The Agent tab goes further than search: it can read and write in the vault too, sandboxed to it. - MCP integration lets external AI tools query your vault without copy-pasting. - Routines are available on every plan, including Free. What differs by plan is the monthly allowance of connector calls a routine spends when it reaches Gmail, Outlook, Notion, or web search. - You can switch plans anytime. Downgrading keeps local data; cloud/team features pause. --- ## Download Canonical: https://www.hyprcore.ai/download Native builds for macOS (Apple Silicon and Intel) and Windows x64. Code-signed and auto-updating. The page lists the latest release plus every published version, and offers the build matching your machine first. --- ## Changelog Canonical: https://www.hyprcore.ai/updates ### v1.11 — Hyprcore runs on Windows. And setup proves it works before you start. (2026-09-04, major) For a long time the answer to “is there a Windows version?” was no. Now there is one — dictation, meetings that capture both sides of the call, and screen recording, with transcription still running on your own machine rather than someone else’s. Setup was rebuilt around proof as well: it ends when you have said a sentence and watched your own words come back, not when a download finishes. And your meetings can live in an Obsidian vault you already have, one click away. Highlights: - Run Hyprcore on Windows — dictation, meetings, and screen recording, on Windows 10 and 11 - Record both sides of a call on Windows, with no audio driver to install and nothing to allow - Install the Hyprcore plugin into an Obsidian vault you already have, in one click - Finish setup only once you have said a sentence and seen your own words come back - Stop losing an afternoon to a meeting you forgot to end — the recording notices the call is over New: - Hyprcore runs on Windows — the whole app, not a cut-down one. Hold Ctrl+Space and talk, and the words go into whatever you were typing into. Meetings, action items, the vault and your local models all work the way they do on a Mac. The installer is signed, it updates itself, and it needs Windows 10 (version 1809) or later - Meetings on Windows record both sides of the call: your microphone and the sound coming out of your speakers, kept apart and then mixed, so the other people on the call end up in the transcript instead of just you. There is no audio driver to install, no cable to route, and nothing to allow - A meeting on Windows can capture your screen alongside the audio - Windows never asks an app like Hyprcore for permission to use your microphone — it is a switch in Settings you may have flipped years ago. If it is off, Hyprcore says so and takes you to it, instead of behaving as though dictation is broken - Put your meetings inside Obsidian. Settings → Integrations → Obsidian lists the vaults Obsidian already knows about and installs the Hyprcore plugin into the one you choose, so meetings, notes and attachments arrive among your own notes rather than in a second vault of Hyprcore’s shape. The plugin reads the app on your own machine — no account, no API key, nothing sent anywhere - Open the vault folder in your file browser straight from Settings → About - Setup now ends in proof rather than a progress bar. Four screens, and the last one waits until you have said a sentence and watched it come back — one thing to try that confirms the microphone, a working input device, the download, and transcription all at once. If it will not work after three tries, it hands you a support message with the details already filled in instead of leaving you stuck - One download now covers both dictation and meetings, instead of two - Make the whole interface bigger — text, icons and spacing together — if the default is small on your display - A routine can write down what it found as a note in your vault, and tick off a commitment it can see you have already finished, so a morning briefing stops reading back every promise you have ever made - A guide to routines, with a walkthrough you can replay Improved: - A recording notices when the call is over. Leave a meeting without stopping Hyprcore and it now sees the call end and tells you, instead of running for hours into an empty room - A meeting can take its name from the calendar event it belongs to - Speaker separation stays within the number of people you said were in the room, and leaves a voice unnamed rather than guessing between two equally close matches. It can also tell you how much of a recording is people talking over each other - Removing a meeting’s media can drop the video and keep the audio - Hyprcore opens dark by default now. If you chose the light theme, it stays Fixed: - Speakers swapping places partway through a transcript - Dictation adding sign-offs and subtitle credits that nobody said - A tapped shortcut recording nothing at all, and a short press quietly latching the recording on - Pasted text arriving mangled because the clipboard came back too early - Rebuilding a meeting un-ticking action items someone had already finished - Google Calendar failing to sync, and a sync that would not settle - Signing in looping back on itself, and a hang at startup on a poor connection - Tables written by the assistant collapsing into run-on text - A stranded recording hiding a meeting and inflating your recorded hours - A video turning into a new chapter every two seconds while it played ### v1.10 — Write the instruction once. It runs while you are away. (2026-08-14, major) Until now Hyprcore only worked while you were sitting in front of it. Routines change that: a trigger, an instruction in plain English, and a list of what it is allowed to touch. A routine can run on a schedule, half an hour before a meeting, or the moment one ends — reading your mail and your notes to brief you, drafting a follow-up for you to look at, or posting a finished meeting to whatever you already automate. Ten of them come already written, and every one of them is a file in your vault you can open and read. Highlights: - Routines — tell Hyprcore what to do in plain English, and when - Connect Gmail, Outlook, Notion and web search, read-only until you say otherwise - Webhooks that POST a finished meeting to n8n, Zapier, Make or your own endpoint - Ask a question from anywhere, by voice or by typing, answered from your vault - Run routines and chat on your own Claude Code subscription instead of ours New: - Routines: a trigger, an instruction written in plain English, and a list of what the routine may reach. It can run on a schedule, a set number of minutes before a calendar event, when a recording starts, when a meeting ends, or only when you press the button - Ten routines come already written — meeting prep before an external call, a morning briefing, a Friday review, a sweep for commitments going stale, follow-up drafts, a research brief, a topic watch, and three that post to your own systems. A template fills in the editor and waits; nothing starts running because you clicked a card - Two separate settings decide how much rope a routine gets. One covers your machine — whether it can touch files and run commands. The other covers change — read only, stop and ask you first, or act on its own. “Draft my replies but never touch my disk” is a normal thing to want, and one slider cannot say it - Connect Gmail, Outlook mail, Notion and web search. The settings panel tells you what a connection can reach in words — read your mail, draft replies — rather than showing you a scope URL and leaving you to work it out - Every tool in a connected account is marked read or write by hand, and anything unrecognised counts as a write until someone has looked at it. The limits you set are applied to the request before it leaves your Mac, not asked of the model as a preference - Webhooks: a routine can POST to any HTTPS endpoint — or to localhost, for the n8n running on your own machine. Signed so the receiver can verify it, retried six times over about eight hours, and you choose what goes in the body: summary, action items, attendees, transcript, notes, links - A delivery log showing every attempt, its result, and the exact body that was sent — plus a test button that posts a real payload built from a real meeting before you turn anything on - A run log that accounts for every run: what fired it, what it called, what it produced, and what it cost - Ask a question from anywhere with a shortcut — by voice or by typing — and get an answer from your own meetings and notes. Type @ to point it at a specific note or meeting - If you have Claude Code installed, routines and chat can run on your own subscription instead of Hyprcore’s. The settings panel names which credential it would actually sign in with, because an API key in your environment quietly outranks a subscription and would otherwise bill the wrong account - PDFs open in the app with a real page view, images come into your vault as files in their own right, and the assistant can look at either — asking you first, every time - Support moved into the app, and it arrives already knowing your version, your platform, and what went wrong Improved: - The meeting page was rebuilt — a new header, clearer tabs, and a transcript panel that is easier to read and move around in - A progress bar while a meeting is being made, showing which step is running and which are still to come, instead of an unexplained wait - The Usage tab says which routine spent what, so a routine quietly working through your monthly allowance is visible rather than deduced from a bill - Import a file straight into the folder you are looking at, from the folder’s own menu Fixed: - Renaming a speaker not showing up until you reopened the meeting - Renaming a meeting leaving its own record pointing at the old folder - Notes written by the assistant using formatting the note editor could not display - Images not appearing in the import dialog - Spacing in chat answers, and @-tagging not working in the in-app composer ### v1.9 — Who said what, sorted. And you can send the video now. (2026-08-10, major) Speaker separation was rebuilt from the model up, and it is the difference between a transcript you trust and one you have to re-listen to. Where it used to confuse one voice for another in roughly one moment out of five, it now does so in one out of ten — and on a long meeting it works several times faster than it did. Shared meetings carry the video as well, with chapters and a download, and you can pick a different transcription model for a single meeting without changing your default. Highlights: - Speaker separation is roughly twice as accurate, and much faster on long meetings - A shared meeting now plays the video, with chapters and a download - Pick a different transcription model for one meeting, without touching your default - Hyprcore tells you when it loses the room audio instead of recording silence - An interrupted model download resumes instead of leaving you unable to transcribe New: - Shared meetings play as video, not just audio and a wall of text — with chapters to jump between and a download for the people who want the file - A compact model picker in the start-meeting dialog: choose the transcription model for this one meeting and leave your default alone - Hyprcore now warns you mid-recording if it stops hearing the room, both in the app and in the small pill, rather than quietly recording nothing - Reveal in Finder for a meeting’s raw audio and video, so the original files are one click away - Transcripts made in the cloud now carry real timestamps and word-level timing, the way local ones already did Improved: - Speaker separation was rebuilt on a new voice model and now compares overlapping windows and votes on the result, instead of deciding once and living with it — it confuses one speaker for another roughly half as often as before - Long meetings separate speakers several times faster, so a two-hour recording is not the wait it used to be - Your microphone and the room audio are levelled once, evenly, rather than being ridden up and down frame by frame — quiet speakers stay audible and nobody swells - System audio is captured correctly whatever your device is set to, so a recording at 44.1kHz no longer plays back at the wrong speed and pitch - Signing in on the desktop hands the session over directly instead of asking you to click one more button Fixed: - A summary, title, or set of action items coming back in a different language than the one the meeting was held in - Model output arriving wrapped in stray code fences, and long documents being mangled by the attempt to strip them - The meeting picker hanging, and a transcript disappearing without a word - A dead-end screen when asking for audio and video on macOS 14, which cannot do it — you are now told up front - A meeting’s per-speaker audio losing track of who was on which side - The capture dialog opening partly off the edge of the window ### v1.8 — Take the tour, then hear every speaker for who they are. (2026-08-05, major) A guided tour now walks you through Hyprcore and asks you to try dictation for real, in a box on the screen, rather than telling you it works. Underneath it, the biggest pass we have made on meeting transcripts: the echo of your own speakers no longer comes back as a second speaker, remote participants stop being renamed to you, and long silences stop turning into invented sentences. And the whole cloud model lineup is current again — twelve new models to choose from, with DeepSeek V4 Flash and Grok STT taking over as the defaults. Highlights: - A product tour that has you actually dictate, not just read about it - Seven new cloud models for summaries and dictation, five new ones for transcription - DeepSeek V4 Flash is the new default — quicker and cheaper than what it replaces - Your own speakers no longer echo back into the transcript as a second person - Remote speakers keep their own identity instead of being labelled as you New: - A product tour — a short guided walkthrough of the app, with chapters you can take one at a time and skip whenever you like - The dictation step in the tour hands you a text box and waits for you to talk into it, so you finish knowing it works on your machine rather than being told - Settings → Models → Local now shows whether your local runtime is up, and links you to it when it is not — download buttons wait until it can answer - Seven new cloud models for meeting summaries, chat, and dictation clean-up: DeepSeek V4 Flash and V4 Pro, GPT-5.6 Luna and Luna Pro, MiniMax M3, GLM-5.2, and Gemini 3 Flash - DeepSeek V4 Flash becomes the cloud default — faster and cheaper than the model it replaces, and it moves you over once. If you picked a model by hand, yours stays exactly where you put it - Five new transcription models: Grok STT, Deepgram Nova-3, Qwen3 ASR Flash, GPT-4o mini Transcribe, and Fish Audio Transcribe - Grok STT becomes the default for meetings, and Qwen3 ASR Flash is the one to reach for if you work in Chinese - The Speech-to-Text tab now tells you which models are actually in use, the way the Intelligence tab already did Improved: - Meeting transcripts are cleaner end to end: pauses are no longer swallowed into the sentence before them, and a long block of speech is broken into the turns it really was - Transcripts made on your own machine now carry word-by-word timing, so speaker changes are cut at real word boundaries instead of handed wholesale to whoever talked most - Speaker separation re-thinks its own decisions instead of committing to the first grouping it finds, so one bad moment of audio no longer drags a whole speaker along with it - Meetings cost noticeably less to run in the cloud — the same audio is no longer transcribed and billed twice, and long meetings stop paying for a summary rewrite every fifty seconds - Importing an existing library of recordings works properly: speakers are separated, and transcripts from other apps land with their timings intact - A machine that was reinstalled or reset no longer holds two of your device slots forever Fixed: - The same sentence appearing twice in a meeting — once as you, once as the person actually speaking — in both the live view and the finished transcript - A remote participant being renamed to you because your microphone happened to be picking up the room - Long silences coming back as invented sentences, sometimes in a language nobody on the call was speaking - Chunks of long meetings being dropped from cloud transcription without a word about it - Dictating raising the "meeting detected" banner, because Hyprcore heard itself holding the microphone - The Ask panel on a saved capture rendering off the bottom of the screen with no way to scroll to it - Two captures with the same filename in different folders sharing one conversation ### v1.7.0 — Speak in your language. Teach Hyprcore your words. (2026-07-25, major) Pick the language you are speaking and Hyprcore keeps it that way — or turn on translation when English is what you need. Your custom vocabulary now follows you everywhere, so names, acronyms, and product terms come out right in both dictation and meetings. Highlights: - Choose the language you are speaking, or translate it to English when you need to - See exactly which languages each transcription model supports - One shared vocabulary for dictation and meetings - Names, acronyms, and product terms are corrected consistently across your vault - Cleaner, more dependable speaker labels in meeting transcripts New: - Choose the source language for dictation and meetings, with an optional Translate to English switch - A language-support guide on every transcription model, so you can choose one that fits the way you speak - A shared Vocabulary in Personal settings — add names, acronyms, and product terms once, then use them everywhere Improved: - Custom words now improve meeting transcripts as well as dictation - Speaker labels and diarization hold together more reliably on real conversations Fixed: - Settings that could quietly force English even when you had chosen another language - Models treating automatic language detection as English ### v1.6.5 — Take your meeting with you. (2026-07-24, major) You can now download any meeting as Markdown, plain text, or Word — from Hyprcore or from a shared link. It is also easier to put a real name to each voice, so your transcript becomes more useful over time. Highlights: - Download meetings as Markdown, plain text, or Word documents - Download from a shared meeting page too - Review and name speakers after a meeting - See transcription progress more clearly while a meeting is being processed - More reliable sharing, titles, and speaker suggestions New: - Download any meeting in Markdown, plain text, or Word format - Download shared meetings without asking the sender for another export - A speaker review flow that lets you match a voice to the person who spoke Improved: - Speaker suggestions carry more context from your meeting and calendar - Meeting processing makes its current status easier to understand Fixed: - Share links, notes, titles, and speaker tools all find the meeting they belong to again ### v1.6.4 — Your calendar is now part of the meeting. (2026-07-23, major) Connect Google Calendar or Microsoft Calendar and Hyprcore brings your upcoming meetings, attendees, and call links into one place. Meetings start with the right context, and it is easier to recognize who is speaking once the call is over. Highlights: - Connect Google Calendar or Microsoft Calendar - See upcoming meetings, attendees, and conference links in Hyprcore - Bring calendar attendees into speaker review and your people records - Keep screen recordings safe if a capture is interrupted - Meeting settings no longer open browsers just by checking their availability New: - Google Calendar and Microsoft Calendar integrations, with upcoming events, attendee names, response status, and meeting links - Calendar attendees become people you can recognize and use when reviewing speakers - Per-word timing in supported transcripts, for more precise speaker attribution Improved: - Speaker matching uses the people and attendees already connected to the meeting - The app is more resilient when a screen capture stops unexpectedly Fixed: - Opening Meeting settings no longer launches every installed browser ### v1.6.3 — Bring your files in. Keep the context. (2026-07-21, minor) Bring in audio, documents, and email exports in batches, then ask Hyprcore about them alongside your meetings. Home is also calmer and faster to navigate, with clearer ways to start a meeting or import what you already have. Highlights: - Import audio, documents, and email exports in batches - Turn imported material into context Hyprcore can search and answer from - A clearer Home with dedicated Start a meeting and Import files actions - More honest model and provider settings - More reliable cloud sync and meeting sharing New: - Import batches of audio, documents, and email exports into your vault - Ask questions about imported material alongside your existing meetings and notes - A refreshed Home with quick ways to start a meeting or import files Improved: - Model and provider settings now make it clear what will actually run - Cloud sync follows folder changes and sharing is easier to reach Fixed: - A release issue that could stop the macOS installer from being published ### v1.5 — Hit record, talk to yourself, get a document back. (2026-07-11, major) Press record, walk through whatever is on your screen, and Hyprcore hands back a note with a player, a click-to-seek transcript, chapters, and a written document — no meeting, no second person required. Meetings now summarize themselves while they are still running. And Hyprcore will finally spell your teammates' names and your product's jargon the way you spell them, because you can teach it. Highlights: - Recording notes — record a walkthrough on your own and get back a transcript, chapters, and a written doc - A summary that writes itself while the meeting is still happening - A custom vocabulary, so names, acronyms, and product terms come out right - An Agent tab that can search your vault and act in it, not just talk about it - Onboarding down to three steps, with the model downloading in the background New: - Recording notes — record a screen-and-mic or audio-only walkthrough, and it lands in your vault as a note with a media player, a transcript you can click to jump around, chapters, and an AI-written document - Ask questions about any recording, the same way you already do with meetings - A quick-record shortcut you can bind, plus a way in from Home, the sidebar, the command palette, and any folder - Live meeting summary — the Summary tab fills itself in while you are still on the call, instead of making you wait for the end - Custom vocabulary — add names, product terms, and acronyms in Settings, and every transcription uses them: local or cloud, live or after the fact - An Agent tab on the Knowledge page — a chat that can actually search, read, and write in your vault, inside a sandbox that keeps it there - A raw / enhanced switch on summaries, for when you would rather see what was really said - A theme picker, so the app can look the way you want it to - Credits for managed cloud transcription and summaries — you pay for what you use, and nothing more Improved: - Onboarding is three steps now — welcome, permissions, one model — and that model downloads in the background while you get on with it - The recording-consent notice appears at your first recording, where it means something, instead of during setup - A meeting model is offered the moment you need one, right inside the record dialog - A microphone left too quiet is detected and boosted for the recording, then put back exactly as you had it - Recordings link themselves to the calendar event they belong to - Failures during post-processing tell you what happened instead of dying quietly in a log file Fixed: - Transcripts inventing sentences during silence, and one speaker being credited with words from the other side of the call - Live transcription cutting out mid-meeting on some cloud connections - A dictation failure that showed nothing at all on the overlay - Meetings that looked like they had no model when one was sitting there, ready ### v1.3 — Say “Hyper,” then say what you want. (2026-06-16, major) Action Mode turns your voice into instructions instead of text: say “Hyper,” tell it to shorten this, rewrite that, or answer this email, and it does it — hands never leaving the page. Alongside it, a companion panel that sits beside your browser and pulls what you are reading into your vault: save a page, get it summarized, and ask it questions. Highlights: - Action Mode — say “Hyper” and an instruction, and Hyprcore carries it out - A companion panel that saves the page you are reading straight into your vault - Read and summarize any web page or YouTube video in one click - Ask questions about anything you have saved, right there on the page - Unread dots, so you can see what you have not gotten to yet New: - Action Mode — open with “Hyper” (“hey Hyper” and “ok Hyper” work too), then say what you want, and your words run as an instruction instead of being typed out - Action Mode reads whatever text you have selected as context, so “make this shorter” means this - Its own hotkey, its own settings tab, and your own wake words if “Hyper” is not for you - The companion panel — a floating window beside your browser for saving, summarizing, and highlighting what you read - Save any web page or YouTube video into your vault as a plain note you own - Read & summarize — one click captures the page and writes you a summary of it - Ask — chat with any saved page or video, either on the capture itself or from the companion panel - Send a highlight straight to your vault: select the text in your browser, hit save - Summary, Content, and Notes tabs on every capture, with notes you can actually edit - A tabbed Templates page, with a live preview of what Action Mode will do Improved: - Regenerate a summary in place with instructions — “shorter,” “just the numbers” — instead of starting a new note - Unread dots on saved captures, cleared as you read them - Speech recognition now understands the ways “Hyper” actually comes out of a microphone — “hiper,” “high per,” and the rest Fixed: - A crash when asking questions about long pages in Chinese, Japanese, or Korean - Empty YouTube captions being saved as an empty note - The Capture dialog refusing to open from the command palette on some screens ### v1.2 — Watch your words appear as you say them. (2026-06-10, major) Dictation no longer waits for you to finish. Words stream onto the page phrase by phrase as you speak, behind a completely rebuilt overlay — a compact glass capsule with a live voice meter that never steals focus from whatever you were typing into. And a proper home for who you are: Personal and Workspace settings, with real avatars. Highlights: - Live dictation — words land as you speak instead of after you stop - A rebuilt recording overlay: compact, glass, with a voice meter that moves with you - The overlay stays out of the way and never takes focus from the app you are typing into - Personal and Workspace settings, with avatars you can generate, upload, or leave as initials - A calmer, cleaner calendar New: - Streaming dictation — your words arrive in the app as you speak them, phrase by phrase, instead of all at once at the end - A rebuilt dictation overlay — a compact glass capsule with a live voice equalizer that opens and closes with the recording - Personal settings — set your name and pick an avatar: generate one, upload one, or keep your initials - Workspace settings — give your workspace a name and an identity, stored as plain files in your vault like everything else - A refreshed calendar across both the week and day views Improved: - Dictation listens only when you are actually speaking, so fan noise and room hum stop turning into words - Streaming stays off until you turn it on — classic dictation is exactly where you left it - Changes you make in Settings are written straight to your vault, not buried in an app database Fixed: - The overlay stealing keyboard focus from the app you were dictating into - A second Hyprcore icon appearing in the Dock whenever the overlay opened - A flash of focus on the first dictation of every session - Words repeating themselves while streaming ### v1.1 — Your calendar, your AI tools, and a command line of your own. (2026-06-04, major) Your week now lives on the Home screen, and Hyprcore can record and join the meetings you pick all on its own. Connect Hyprcore to Claude, Codex, Cursor, and other AI tools in a single click. And for anyone who lives in the terminal, there's now a read-only hyprcore command line that reads straight from your vault. Highlights: - A week-and-day calendar right on your Home screen - Auto-record — Hyprcore joins and captures the meetings you choose, hands-free - One-click connections to Claude, Codex, Cursor, and other AI tools - A new hyprcore command line for reading your vault from the terminal - Model downloads that resume instead of starting over New: - Home calendar with week and day views — your schedule sits on the Home screen, with no second app to open - Record and join straight from any calendar event, with controls right on the event - Auto-record toggle — choose which meetings Hyprcore joins and captures on its own - Integrations settings with one-click setup for Claude Desktop, Codex CLI, Cursor, and other AI tools - A copy-paste setup guide for any AI tool not yet on the one-click list - A built-in connection test, so you know a tool is wired up before your next call - The hyprcore command line — eight read-only commands to list meetings, read transcripts, and search your vault from the terminal Improved: - Interrupted model downloads now resume where they left off instead of starting from zero - The connector and command line always read from the account you are signed into - Sign-in sessions refresh quietly in the background, so you stay logged in Fixed: - A stalled model download that could loop forever instead of finishing the last piece ### v1.0 — Your vault, in plain files. (2026-05-22, major) Hyprcore 1.0. Everything the app knows about your meetings — transcripts, summaries, action items, notes, people, templates, even chat history — now lives as plain Markdown, YAML, and text files on your own disk. No hidden database. Open them in any editor, search them in an instant, back them up like any other folder, and take them anywhere. On that foundation: a cleaner meeting page, real full-text search, and the first way to bring your meetings into other AI tools. Highlights: - Every meeting, note, person, and template stored as plain files you can read, edit, and back up - A meeting page split into Summary, Action Items, and Notes tabs - Instant full-text search across your entire vault - A connector that brings your meetings into Claude and other AI tools - Folders that work like pages, with rename-in-place and starter folders out of the box New: - Plain-file vault — transcripts, summaries, action items, notes, people, templates, and chat history all saved as Markdown, YAML, and JSONL on your disk; the internal database is gone - Full-text search across every meeting and note, with results as you type - A standalone Hyprcore connector that lets Claude and other AI tools read your vault, set up in one click - People pages — name a speaker once and they get their own file, linked into every meeting with [[wikilinks]] so the same person stays connected across sessions - Meeting pages split into Summary, Action Items, and Notes tabs - A sticky, draggable recording mini-player you can move out of the way - Folders are pages too — switch between a written page and the files inside, and rename folders or notes right from the sidebar - Starter Work, Personal, and Archive folders seeded for every new vault - A Licenses screen listing the open-source software Hyprcore is built on Improved: - Cover images and icons now stick reliably to folders, meetings, and notes - A bounded media card replaces the old full-bleed meeting video - Free plans now include a team and a device seat, set up automatically on first sign-in - Billing charges fractional credits instead of rounding every request up to a whole one Fixed: - Edits no longer get lost when switching between meeting tabs - Sidebar menus no longer get clipped inside the sidebar - A note's title is no longer tied to its first heading, so renaming behaves the way you expect ### v0.9 — Shareable meeting links, in-app support, and a clearer privacy choice. (2026-05-09, minor) Three small releases across a few days (0.9.0–0.9.2): send anyone a read-only link to a meeting, reach support without leaving the app, and make an explicit choice about anonymous analytics right in onboarding. New: - Shareable meeting links — send a read-only web view of any transcript or summary, sourced from your Markdown and revocable any time (0.9.0) - A privacy step in onboarding with a clear Off / Errors only / Full choice, and a matching Privacy section in Settings (0.9.0) - In-app support chat and a help widget, so you can reach us without leaving Hyprcore (0.9.2) Improved: - Steadier system-audio capture, with better diagnostics when macOS screen recording misbehaves (0.9.1) - Anonymous product analytics are strictly opt-in, tied to the privacy choice you make in onboarding (0.9.1) - The web app picks up proper favicons and accurate per-page titles (0.9.2) ### v0.8 — New speech engines, a menubar mini-bar, and a quieter, more honest app. (2026-05-05, major) The 0.8 line shipped across a week and a half. NVIDIA's Canary speech engines joined the lineup with built-in translation, the menubar got a real mini-bar with a glass recording pill, the macOS permission check became something the app actually trusts, and we turned on opt-out crash reporting so we can fix bugs before you have to write in. Plus a stack of fixes for testers on Sequoia 15.7.2. Highlights: - NVIDIA Canary speech engines — fast, multilingual, and translate while you talk - Menubar mini-bar with a glass recording pill, stop / cancel, and a live waveform - Anonymous crash reporting on by default — flip it off in one click in Settings - Live transcription that survives any language, including Hindi, Chinese, and Japanese - Permission buttons that respond instantly, without the ten-second pause New: - NVIDIA Canary local speech engines, with built-in translation between supported languages - Menubar mini-bar — start dictation, stop a meeting, or jump back to the app from anywhere on your Mac - Recording pill is now a roomy 340×52 glass panel with stop, cancel, and a live 25-bar waveform - Anonymous crash reporting so we can find and fix bugs faster, with nothing about your meetings or notes attached - New Privacy section in Settings to switch reporting between Off, Errors only, and Full - Bundled macOS permission helper that reads real Microphone, Accessibility, Screen Recording, and Audio Capture status - Active speech model badge in the sidebar is now a one-click model switcher - Permissions panel embedded in Meeting Capture and Dictation settings — fix a denied permission without hunting for it - Stop-recording button right in the meeting topbar, with a clear error dialog if something goes wrong - Built-in scrubbing for file paths, keys, and personal data before anything ever leaves your Mac Improved: - Sidebar, topbar, page header, and buttons all picked up the v0.7.0 design polish - Auto-updater shows real download progress instead of a placeholder bar - Diarization models live in a shared place on disk; old downloads migrate automatically the first time you open the app - Onboarding for the local Gemma model and the Accessibility step both cleaned up - Onboarding handles quick restarts and multiple sign-ins more gracefully - Vault breadcrumbs are consistent across list, note, and meeting views - Meeting banners are now a single, unified modal - Translations synced across all 16 languages - Slack huddles are now correctly recognized as meetings Fixed: - First-launch crash on 0.8.9 — the app opens cleanly on 0.8.10 - Live transcription no longer cuts out mid-meeting on non-Latin speech - Permission buttons in onboarding open the macOS prompt immediately instead of after a long pause - Rapid clicks on a permission button no longer make the app feel laggy - Memory leak when re-downloading the Parakeet speech model - Crash when clicking the menu bar icon before finishing onboarding - Onboarding sometimes restarted after sign-in — it now remembers it is done - A crash that could appear right after sign-in on a fresh install - Removed the experimental Apple Intelligence integration that was triggering a launch crash on some Macs - Cloud sync no longer fails with a header parsing error when credentials are pasted with stray whitespace - Share button on a meeting is disabled until cloud sync is signed in and turned on, matching how it actually works - A meeting-recording crash that could hit when starting a second meeting in a row ### v0.7.0 — A new look, a floating recorder, and a vault that feels like home. (2026-04-28, major) Hyprcore's biggest release yet. The whole app has been rebuilt on a calm, document-style design system. Meetings now record in their own floating window, the vault feels like a real workspace, ⌘K opens anywhere, and notes get Notion-style icons and cover images. Plus in-app analytics opt-in and a refreshed brand. Highlights: - A full visual revamp — warm neutrals, terracotta accent, document-style editor - A floating meeting recorder window with live transcript, notes, and ask in one place - Notion-style page icons and cover images for notes, folders, and meetings - ⌘K command palette with real search and recency — jump anywhere, instantly - A new Home screen, two-pane Settings, and revamped Models, Intelligence, and Subscription pages New: - Floating meeting recorder window — drag the pill, expand for live transcript / notes / ask, traffic-light controls (red stop, yellow collapse), pre-warmed at startup so it appears instantly - Page chrome — pick an icon and a cover image for any note, folder, or meeting; 20 bundled gradient covers, curated icon set, per-page cover repositioner - ⌘K command palette — real-time title search, recents, and base actions (start meeting, dictation, new note); click the sidebar search bar to open it - Home screen — quick actions, today, recently opened, and open loops in one place - Settings rebuilt as a two-pane layout with grouped navigation; new Updates, Subscription, Sync, Profile, and Meetings-permissions pages - Meeting note rewritten as a document with the transcript drawer beside it; video gets a hero embed under the meta - Usermaven analytics opt-in — tracks signup, login, upgrade, and team invite - Onboarding consent step — explicit recording-notice screen before model selection - Notion-style folder management in the sidebar — rename, delete, new note, new subfolder, pick an icon - Default Work, Personal, and Archive folders seeded for new vaults - Topbar processing pill and ghost placeholders while a meeting is post-processing - Live recording timer and inline Stop button on the meeting note itself - Global shortcuts for "start meeting" and "new note" - Meeting templates — manage templates and edit the prompts the AI uses to summarize - Per-card Download button when an Ollama model is not yet installed - Onboarding lets you pick a Gemma variant with size and spec hints Improved: - Refreshed brand — new Hyprcore icon, wordmark, and application icons across every platform - Sidebar is now a nested tree with a workspace switcher and the real vault hierarchy (no more demo data) - Note editor uses serif headings and a document-style layout that reads better at length - Models page revamped with engine chips, metadata cards, and clearer provider management - Intelligence templates use a grid plus toggle groups for tone and length - Subscription screen now shows usage, team seats, and sharing in a single view - Onboarding revamped with a card layout and clearer step indicators - Streaming transcripts feel more responsive — STT chunk target dropped from 20s to 10s - Tray icon swaps no longer hitch the UI thread; meeting + dictation states are arbitrated through one path - Meetings open from the Vault as a single canonical note row instead of a split note + session pair Fixed: - Recorder pill now drags reliably on macOS NSPanel without losing the first click - Live transcript backfills correctly when reopening a session that was already running - Vault de-duplicates meeting folder names that would otherwise collide on the same parent - Cover gallery and icon picker preload correctly when a page first opens - Stripe API version no longer pinned, so the web build survives minor SDK bumps ### v0.6.1 — Grok joins the lineup. (2026-04-19, patch) xAI's Grok lands in the cloud model picker alongside a few smaller models worth having around. Free users get a clearer path to upgrade, and the auto-updater now actually ships the artifacts it is supposed to. New: - Grok 4 Fast and Grok Code Fast in the cloud AI picker - Gemini 2.5 Flash Lite, Ministral 8B, and Llama 3.3 70B added to the catalog - Free-plan cloud option now shows an Upgrade button that opens the upgrade page in your browser Improved: - Cloud credit pool protected: the managed tier now tops out at Haiku-class models so nobody blows the shared budget on a single run - Default web app host is now app.hyprcore.ai everywhere — auth, entitlements, settings Fixed: - Auto-updater bundles (.app.tar.gz + signature) now actually ship alongside the DMG, so v0.6.0 installs can update themselves to v0.6.1 without a manual reinstall ### v0.6.0 — Updates, delivered quietly. (2026-04-18, major) Hyprcore now updates itself. A quiet background check on every launch, a small toast when a new version is ready, and a dedicated Updates tab in Settings when you want to see what's coming. Highlights: - Background auto-update check every time the app launches - A subtle in-app toast when a new version is ready - A new Updates tab with release notes, channels, and install progress New: - Signed, automatic in-app updates powered by a dedicated release channel - Update notification toast — dismiss it or install on the spot - Updates tab in Settings: release notes, progress bar, Remind later, plus a channel selector (beta channel coming soon) - Every release is cryptographically signed so the app only installs builds you can trust Improved: - Release channel is faster and more reliable — you'll see new versions within minutes of them going out ### v0.5.0 — Action items, and Hyprcore for teams. (2026-04-10, major) Meetings become documents you can edit, check off, and reorder — right next to the transcript. Plus a new Pro Team tier so a whole team can share a plan, with per-member device allowances and email invites. Highlights: - A rich notes editor and action items inside every meeting - Pro Team tier with per-member devices and email invites - Everything still saved as portable markdown you can take anywhere New: - Pro Team tier — one plan, multiple teammates, per-member device allowance, and email invites handled in-app - Rich text editor (Tiptap) inside every meeting — write, format, and attach notes to the transcript - Action items with check, edit, reorder, and quick-assign - Drag-and-drop ordering for any list inside meeting notes - Expanded cloud AI model list with more supported languages - Folder-level locking when a vault folder is in use to prevent accidental edits mid-recording Improved: - Meeting pages load noticeably faster on vaults with 200+ sessions - Summary blocks render immediately from cache, then stream updates as AI finishes - Cloud model selection is smarter about picking a default for the language you are speaking - Action item keyboard shortcuts now mirror standard todo app conventions Fixed: - Opening `meeting.md` now loads the in-vault editor with summary and action items instead of a raw file view - Rare case where the action-item list would reset position after a cloud sync - Crash when opening a meeting while another one was still finalizing transcripts ### v0.4.3 — Diarization tuning and Windows mic fixes. (2026-03-28, patch) Smoother speaker separation on noisy calls, plus a round of Windows-specific audio fixes. Improved: - Diarization is less likely to merge two quiet speakers into one - Speaker labels now survive a mid-meeting app restart Fixed: - Windows: certain USB microphones failed to appear in the device picker - Occasional audio dropout when switching output devices during a recording ### v0.4.2 — Paste reliability, Core Audio tap polish. (2026-03-14, patch) Smaller, sharper release focused on the two things people noticed most. Fixed: - Paste-after-dictation occasionally inserted into the wrong field on multi-monitor setups - macOS system audio tap would briefly clip during the first second of a recording - Menu bar icon stayed in "recording" state if the app was killed mid-session ### v0.4.1 — Speaker labels that stick around. (2026-02-24, patch) Once you rename a speaker, that name now persists across sessions with the same person. New: - Rename a speaker once — the label is remembered the next time they show up Improved: - Speaker confidence indicator in the transcript view when a voice is ambiguous ### v0.4.0 — Cleaner audio. Sharper speakers. (2026-02-09, major) A rebuilt audio pipeline with neural echo cancellation and per-speaker diarization. Meetings in loud rooms, on speakerphone, or with overlapping voices now come out legible. Highlights: - Neural acoustic echo cancellation replaces the old spectral model - Built-in speaker diarization — who said what, automatically - Seven-layer anti-hallucination filter on transcripts New: - Neural AEC powered by a two-stage DTLN model — dramatically reduces echo and background noise - Speaker diarization with live labels in the transcript as the meeting unfolds - Anti-hallucination filter that catches the seven most common Whisper misfires (phantom "thank you", repeated phrases, silent-period drift) - Per-segment Whisper timestamps instead of coarse per-chunk timing Improved: - Transcripts in very quiet segments are noticeably cleaner - Mic-mute toggle during recording no longer drops a partial word Fixed: - Cloud diarization provider: retry logic on transient network failures - Crash when switching audio devices during the first five seconds of a recording ### v0.3.2 — Hallucination filter and a full i18n refresh. (2026-01-20, patch) Plus translations completed for the meeting-vault interface in every supported language. New: - Early version of the anti-hallucination filter available as a toggle Improved: - All meeting-vault and action-item strings now localized across every supported language - Startup time faster on first launch after an update Fixed: - Non-Latin characters in meeting titles caused a render glitch in the sidebar ### v0.3.1 — More cloud models and batch Deepgram. (2025-12-18, patch) Room to choose the right model for the job, without paying per-minute where you do not need to. New: - Expanded list of cloud speech models, now with per-language recommendations - Deepgram batch mode for long recordings — cheaper, still accurate Improved: - Cloud/local model switcher remembers your last choice per meeting type ### v0.3.0 — Meetings are a first-class thing. (2025-11-24, major) Hyprcore is no longer just a dictation tool. It records, transcribes, and organizes meetings in a dedicated vault — with summaries, markdown storage, and a session view you actually want to open. Highlights: - A meeting vault with dedicated folders for every session - Live transcripts with a session detail page - Every meeting stored as markdown you can open in any editor New: - Meeting vault with per-session folders created the moment you hit record - Session detail view — live transcript, speaker cards, and AI summary in one place - Markdown-first storage: every meeting lives on disk as a readable `.md` file with YAML frontmatter - AI-generated meeting summaries with customizable length and tone - Start a meeting straight from the macOS menu bar — no need to open the main window Improved: - Sidebar navigation redesigned around the vault concept - Recording indicator in the system tray is clearer about what mode you are in (dictation vs. meeting) ### v0.2.3 — Keycode and paste fixes. (2025-08-11, patch) The unglamorous but important round of shortcut and clipboard reliability fixes. Fixed: - Non-US keyboard layouts: certain modifier combos failed to register as a push-to-talk trigger - Paste sometimes inserted an extra trailing space on macOS 14+ - Linux: keycode mapping mismatch on Wayland when using a tiling window manager ### v0.2.2 — Small quality-of-life touches. (2025-07-28, patch) A collection of tiny fixes and interactions that should have been there from day one. Improved: - System tray tooltip shows the currently selected model at a glance - Model switcher now auto-focuses the search field when opened - Onboarding window remembers progress if you close it mid-setup Fixed: - Audio device picker occasionally displayed devices twice ### v0.2.1 — Stability on slow disks. (2025-07-14, patch) If your drive is slow or nearly full, your models should still load. Fixed: - Model load failure on slow HDDs or drives with less than 5 GB free - Rare race condition on first launch when multiple models initialized simultaneously - Windows installer: completion hang when running under certain antivirus software ### v0.2.0 — Design refresh and on-demand mic. (2025-06-30, major) A fresh look built on shadcn/ui, smarter microphone management that stops listening when you are not talking, and quality-of-life improvements for multi-user machines. Highlights: - On-demand microphone activation — no more always-on listening - Design system refresh across the entire app - Per-user data directories so families and shared Macs work cleanly New: - On-demand microphone activation — mic turns on only when you ask for it - User-scoped local data directories for multi-user machines - Cloud AI proxy with an explicit allow-list of models and languages - Dynamic window resizing for onboarding and authentication on macOS Improved: - Design system refresh built on shadcn/ui — consistent spacing, type, and controls - Voice Activity Detection is less aggressive about cutting you off mid-sentence - Keyboard shortcut picker is easier to discover and change Fixed: - Recording indicator stuck on after an unclean shutdown - Occasional crash when switching between Whisper and Parakeet models ### v0.1.3 — Linux Wayland crash fixes. (2025-06-09, patch) A Linux-focused patch. macOS and Windows users can skip this one. Fixed: - Wayland: crash when the app tried to read global shortcut events on GNOME 46 - Linux: clipboard injection failed on wl-copy-based desktops ### v0.1.0 — Hello, world. (2025-05-16, major) The first public release of Hyprcore. A local, private, keyboard-driven dictation app that runs entirely on your machine. Highlights: - Local speech-to-text with Whisper and Parakeet - Global push-to-talk shortcut that works anywhere - macOS, Windows, and Linux from day one New: - Core speech-to-text transcription with local Whisper and Parakeet models - Voice Activity Detection to automatically filter silent gaps - Global keyboard shortcuts with push-to-talk support - System tray integration with clear recording indicators - Clipboard integration — transcribed text lands where your cursor is - Multi-model selection: Whisper Small, Medium, Turbo, Large, and Parakeet V3 - Cross-platform from day one — macOS, Windows, and Linux - Automatic updates --- ## Roadmap Canonical: https://www.hyprcore.ai/roadmap ### In Progress - **Graph-based LLM wiki for notes** (Knowledge): Every note, meeting, person, and topic linked in a knowledge graph an LLM can traverse. Backlinks, related notes, and semantic clusters surface automatically as you write. - **Beta release channel** (Platform · Q2 2026): An opt-in beta track inside the new Updates settings tab so early testers can try the next release a few days ahead of everyone else. - **Structured memory graph** (Knowledge · Q3 2026): People, topics, and commitments pulled out of every meeting and stored in a queryable graph. Foundation for cross-meeting search. - **Cross-meeting AI search** (Knowledge · Q3 2026): Ask a question once, get an answer drawn from every meeting you’ve ever recorded. Sources always link back to the original transcript. - **Recorder ask, with vault context** (Knowledge · Q3 2026): The Ask tab in the floating recorder grows beyond the current session — pull in related notes and past meetings so live answers are grounded in your whole vault. ### Planned - **Voice memo import** (Meetings · Q3 2026): Drop an audio file into a watched folder and Hyprcore transcribes, summarizes, and files it automatically. - **Mobile capture companion** (Platform · Q4 2026): A lightweight iOS and Android app that records audio in the field and syncs back to the desktop vault for transcription and summarization. - **AI-assisted templates** (Meetings · Q3 2026): Describe what you want captured in plain language and Hyprcore drafts the template for you — for meeting summaries and action-item extraction alike. - **Custom insight extractors** (Knowledge · Q3 2026): Define the structured details you care about per meeting type — pricing objections from sales calls, strengths from interviews — and have them pulled from every matching meeting automatically. - **Linux meeting beta** (Platform · Q3 2026): The meeting vault, diarization, and echo-cancellation pipeline coming to Linux after the Windows build lands. ### Under Review - **Custom API endpoints for BYOK intelligence models** (Platform): Right now, BYOK only support OpenRouter. I want to use my own endpoints so I've build a local proxy to get it work. It's working now with local server proxy. However, it would be more convinient if I could just have custom endpoints supported from the app - **Incoming Webhooks** (Platform): I use Ringtonic.app. Demand will increase for the use of this tool, the founder is exceptional and it's growing fast. Recordings would be nice to import into the program. - **Meeting Scheduling and Automatic Notifications to Guests You Will Be Recording** (Platform): I noticed that a friend has an automated message come from Fathom that they intend to record the meeting. The email asked for my permission to record. I really like this and wonder if there is a way to incorporate this feature in Hyprcore, especially as geting consent to record in two-way states is important. - **The ability to import data from a spreadsheet** (Platform): Just what it says: import data from a spreadsheet. This way you can import financial data and other key report details, along with datestamps, so you can ask the AI to report on trending / results of actions taken. ### Shipped - **A vault you can read in plain files** (Platform · 2026-05-22): Every meeting, note, person, and template is now stored as plain Markdown, YAML, and text on your own disk — no hidden database. Open them in any editor, search them instantly, and back them up like any folder. - **Obsidian integration** (Integrations · 2026-09-04): Install the Hyprcore plugin into an Obsidian vault you already have, in one click, so meetings and notes arrive among your own notes rather than in a separate vault. - **Instant full-text search** (Knowledge · 2026-05-22): Search across every meeting and note in your vault and get results as you type, powered by a fast on-device index. - **Hyprcore command line** (Integrations · 2026-06-04): A read-only hyprcore command-line tool that lists meetings, reads transcripts, and searches your vault straight from the terminal. - **Shareable meeting links** (Platform · 2026-05-07): Send a read-only web view of any meeting transcript or summary. Revocable, optionally expiring, always sourced from your markdown. Paid share gating with a dedicated viewer. - **Editable meeting templates** (Meetings · 2026-04-20): Manage the prompts the AI uses to summarize each meeting type — adjust tone, length, and section structure, then save the template for reuse. New Intelligence settings with a template grid and toggle groups. - **Durable speaker attribution** (Transcription · 2026-04-14): Name a voice once and keep it labeled forever. A 3-level pipeline (audio stem + ONNX ML + voice enrollment), participant editor with re-diarization, and a voice-profiles store keep speakers consistent across sessions. - **Calm, document-style UI revamp** (Platform · 2026-04-28): A warm neutral and terracotta palette across the whole app, a serif document-style editor, two-pane Settings, and revamped Models, Intelligence, and Subscription pages. - **Floating meeting recorder** (Meetings · 2026-04-28): A dedicated recorder window with live transcript, notes, and ask in one place. Drag the pill anywhere, expand on demand, traffic-light controls — pre-warmed at startup so it appears instantly. - **Page icons and cover images** (Knowledge · 2026-04-28): Notion-style icons and cover images for every note, folder, and meeting. 20 bundled gradient covers, a curated icon set, and a per-page cover repositioner. - **⌘K command palette** (Platform · 2026-04-28): Real-time title search, recently opened, and base actions like start meeting, dictation, and new note — accessible from anywhere in the app. - **Home screen** (Platform · 2026-04-28): A landing surface with quick actions, today, recently opened, and open loops — so the app opens to something useful, not an empty workspace. - **Vault folders, phase 2** (Meetings · 2026-04-28): Rename, delete, create notes and subfolders, and pick an icon — all from the sidebar. Default Work, Personal, and Archive folders seeded for new vaults. - **Quieter auto-updates** (Platform · 2026-04-18): The app now updates itself in the background — signed releases from our own channel, a gentle toast when one is ready, and a dedicated Updates tab in Settings. - **Grok and a wider cloud model catalog** (Transcription · 2026-04-19): xAI Grok 4 Fast, Grok Code Fast, Gemini 2.5 Flash Lite, Ministral 8B, and Llama 3.3 70B now available in the managed cloud tier. - **Hyprcore for teams** (Platform · 2026-04-10): A Pro Team tier with per-member device allowance and in-app email invites. One plan, the whole team, same local-first experience. - **Action items in every meeting** (Meetings · 2026-04-10): Check off, edit, and reorder tasks right inside the transcript view. Saved alongside your notes as portable markdown. - **Rich notes editor** (Meetings · 2026-04-10): Format, link, and annotate meeting transcripts without leaving Hyprcore. Built on Tiptap with drag-to-reorder lists. - **Neural acoustic echo cancellation** (Transcription · 2026-02-09): A two-stage DTLN model replaces the old spectral filter. Calls in loud rooms and on speakerphone come out noticeably cleaner. - **Speaker diarization** (Transcription · 2026-02-09): Live "who said what" labels in the transcript as the meeting unfolds. Works with both cloud and local providers. - **Anti-hallucination transcript filter** (Transcription · 2026-01-20): Catches the seven most common Whisper misfires — phantom "thank you" insertions, repeated phrases, silent-period drift. - **Meeting vault** (Meetings · 2025-11-24): Dedicated per-session folders, session detail pages, and markdown-first storage you can open in any editor. - **Deepgram batch mode** (Transcription · 2025-12-18): A cheaper path for long recordings where real-time isn’t required. Quality stays the same; cost drops meaningfully. - **Menu-bar "start meeting" shortcut** (Platform · 2025-11-24): Kick off a recording from any app without opening Hyprcore’s main window. - **Full interface localization** (Platform · 2026-01-20): Every string in the meeting vault and action-item views is now translated across all supported languages. - **Calendar integration (Gmail, Outlook)** (Integrations · 2026-07-23): Two-way sync with Google Calendar and Microsoft Outlook so meetings arrive pre-labeled with title, attendees, and agenda — and recaps land back next to the event you scheduled. - **Daily and weekly digests** (Knowledge · 2026-08-14): A short, generated brief of what you discussed, what you committed to, and what is still open — assembled from the meetings and notes of the period. - **Connect Hyprcore to your AI tools** (Integrations · 2026-06-04): A standalone connector brings your meetings into Claude Desktop, Codex CLI, Cursor, and other AI tools — set up in one click from the new Integrations settings, with a copy-paste guide for anything not yet on the list. - **Windows app (beta)** (Platform · 2026-09-04): Dictation at full parity, with meeting recording, diarization, and echo cancellation following in beta — the same Hyprcore, now on Windows 11. - **Routines — automatic meeting follow-ups** (Meetings · 2026-08-14): Set an instruction once and Hyprcore runs it after every meeting or on a schedule — digests, follow-up drafts, action-item roll-ups — writing straight into your vault. - **Connect your tools (Gmail, Outlook, Notion)** (Integrations · 2026-08-14): Let a routine reach the accounts you already work in — read your mail to brief you, draft a reply, file a note in Notion — read-only until you grant it more, with per-action approval so nothing leaves without your say-so. - **Outbound webhooks (Zapier, Make, n8n)** (Integrations · 2026-08-14): Fire a webhook from any routine to reach Zapier, Make, n8n, and the long tail of tools — your meetings, wired into whatever you already automate. - **Usage dashboard, with bring-your-own-key** (Platform · 2026-08-14): See where your cloud tokens and connector calls go — broken down by language model, transcription and embeddings, with every routine run costed individually and each connector call traced back to the routine that spent it. Or plug in your own API key, or run chat and routines on your own coding-agent subscription. --- ## Blog Canonical: https://www.hyprcore.ai/blog ### Loom for one: how to turn a five-minute screen recording into a document you can search (2026-07-15) Async video solved the recording problem and left the knowledge problem exactly where it found it — a link nobody can search, skim, or quote. Here is what changes when a solo screen recording comes back as a transcript, chapters, and a written doc in a vault you own. URL: https://www.hyprcore.ai/blog/loom-for-one-recording-to-searchable-document Author: The Hyprcore Team Tags: recordings, async, knowledge, how-to, macOS You know the recording. Someone asks how the billing code works, or how to run the release, or why the staging database is laid out like that — and rather than write a page you will never finish, you hit record, share your screen, and talk for five minutes. It is the fastest honest answer you have. You send the link. They watch it. It works. Then it dies. Six weeks later a new person asks the same question, and nobody remembers the recording existed — and if they did remember, they could not find it, because you cannot search a video. So you record it again. The knowledge was captured and it still evaporated. ## Async video solved the wrong half of the problem Loom and the tools like it made recording effortless. That was a real contribution — it took a thing that used to require a screen-capture app, a file, and a Dropbox link, and turned it into a browser tab. But look closely at what they optimized: distribution. Getting the video from your machine into someone else's eyeballs, fast. That is the recording problem, and it is solved. The knowledge problem is the other one, and nobody touched it. A video link is a dead end for everything you want to do with an explanation after the first viewing: - You cannot search it. Not with ⌘F, not with your notes search, not with anything. The words are in there; they are just not text. - You cannot skim it. A five-minute video takes five minutes — or two and a half at 2×, which is still fifteen times longer than reading the same content. - You cannot quote it. Nobody pastes a video into a pull request to explain why the retry logic is capped at three. - You cannot read it on a train, in a meeting, or anywhere else you cannot play audio. - You cannot hand it to an AI tool. Your assistant can read every document you own and not one second of your video library. - You do not really own it. It lives on someone's servers, behind someone's pricing page, until the day it does not. None of that is a complaint about video. Video was the right way to capture the explanation — you talked through it in five minutes because talking is faster than writing, and pointing at the screen is faster than describing it. The capture was right. The container was wrong. ## What a solo recording should actually come back as Here is the test. You record five minutes about the billing webhook. What should exist tomorrow morning? 1. The recording itself, playable, because sometimes you do need to watch someone point at the thing. 2. A full transcript, timestamped, so the words are text and text is searchable. 3. Chapters, so a person who only needs the last ninety seconds can jump to the last ninety seconds. 4. A written document — the explanation, in prose, that you would have written if you had the hour. 5. All of it in the same place as everything else you know, so it turns up when someone searches for "billing" nine months from now. That last one is the whole game, and it is the one every async-video tool skips. A recording in a separate app with a separate search box is a silo with a nicer player. **This shipped** — Recording notes landed in Hyprcore 1.5. Record your screen and mic — or just audio — and it becomes a note in your vault with a player, a click-to-seek transcript, chapters, and an AI-written doc. Same vault as your meetings and pages. Same search. ## How it works, end to end ### 1. Hit record Start a recording note from the Home screen, the sidebar, the command palette, or a right-click inside any folder — and if you do this often, bind a shortcut to it in Settings so the thought never has to survive a trip through a menu. Two modes: screen and mic, or audio only. Audio-only needs no screen-recording permission at all, which makes it the right pick for thinking out loud on a walk. ### 2. Talk the way you would talk to a person This is the only technique worth knowing: say the structure out loud. "First, where the code lives." "The part that trips everyone up is the webhook." "If a charge fails, here is what you do." You are not performing — you are giving the chaptering something to chapter, and the doc something to organize itself around. Five minutes of narrated clicking produces a better document than fifteen minutes of silent clicking, every time. ### 3. What lands in your vault A directory, next to your meetings and pages, holding plain files: ``` folders/Inbox/2026-07-10_billing-walkthrough_ab12/ ├── index.md # the note: Doc, Chapters, Notes ├── recording-1752...mp4 # the screen + mic recording ├── recording-1752...-mic.wav # the stems, kept └── transcript.jsonl # every segment, timestamped ``` Open index.md in any editor and you will find three sections. Doc is the AI-written explanation, and you can edit it — it is your document now, not a read-only output panel. Chapters is a timestamped outline, one line each, in the shape of "[02:14] The webhook that trips people up". Notes is your own scratchpad, for the thing you remembered ten minutes after you stopped recording. No database. No proprietary format. A folder with a video, a transcript, and a markdown file — which means it backs up like a folder, syncs like a folder, and outlives us like a folder. ### 4. Read it back, three ways Open the note and the player sits at the top with the transcript beside it. Click any line of the transcript and the recording jumps there. Click a chapter and it jumps there. Or ignore the video entirely and read the doc, which is what will happen ninety percent of the time — because reading is faster, and the video was always a delivery mechanism for the words. ### 5. Ask it questions Every recording has its own Ask chat, with the doc, your notes, and the timestamped transcript as context. "What did I say the retry limit was?" is a question you can now ask a five-minute video, which is a sentence that would have been nonsense two years ago. And because the recording is a first-class thing in the vault, it turns up in full-text search alongside your meetings — and the agent can read it when it is answering something else entirely. ## The transcript beats the video, even when the video is good People resist this because the video feels richer, and in the moment it is. But richness is not the axis that matters after the first watch. Reading is roughly four times faster than listening, and skimming is faster still — you can eat a five-minute explanation in forty seconds and know exactly which forty seconds of the video you actually need to watch. Playback speed does not fix this. 2× is still linear. A document is random access. > A video is a queue. A document is an index. The second one is what a team can build on. And the document compounds in a way the video cannot. It gets edited when the answer changes. It gets linked from the page about the billing system. It gets read by the assistant that is drafting the onboarding guide. The video sits still while the document goes to work. ## What this is not This is not a video production tool, and we are not pretending otherwise. There is no editing timeline, no zoom-and-pan cursor choreography, no webcam bubble in the corner, no branded share page with view counts. If you are making a customer-facing demo or a marketing walkthrough, use a tool built for that — you will be happier. There is also, today, no hosted link to send someone. A recording note is a folder on your disk; sharing it means sharing the file, or sharing what came out of it — which for most internal explanations is the doc anyway, and the doc is text you can paste anywhere. If a link is what you need, that is a fair reason to keep using a video tool alongside this. And the doc is AI-written, which means it is a draft with a good memory, not a scribe with perfect judgment. It reorganizes what you said; it does not verify it. The transcript is the record — the doc is the useful lie you get to correct in thirty seconds instead of writing from scratch in thirty minutes. ## The next time someone asks you a question The instinct to record a quick walkthrough instead of writing a page is a good instinct — it is why async video won. The mistake was letting the answer live inside a video file, where none of the things you do with knowledge are possible. **The premise** — Talking is the fastest way to explain something. Reading is the fastest way to understand it later. A recording should give you both — the video for the moment, the document forever. So: the next time someone asks you how something works, hit record and talk for five minutes. Then go look at what is sitting in your vault when you are done — a transcript you can search, chapters someone can skip to, and the page you never had time to write. Hyprcore is free to download and runs natively on macOS and Windows. --- ### Your company's second brain: every meeting, note, and dictation in one vault (2026-06-04) The "second brain" was supposed to hold everything you know. For most teams it holds everything you typed — and quietly loses everything you said. Here is what a second brain looks like when meetings, dictations, and notes share a single local-first vault, and why that changes what your team can remember. URL: https://www.hyprcore.ai/blog/your-companys-second-brain-one-vault Author: The Hyprcore Team Tags: second brain, knowledge, philosophy, product Ten years ago, Tiago Forte gave a name to something a lot of people were already reaching for: a "second brain." The pitch is simple and good. Your biological brain is for having ideas, not storing them. So you build a trusted external system — capture what matters, organize it, and get it back the moment you need it. The book that followed, Building a Second Brain, turned it into a movement, complete with a method (PARA: Projects, Areas, Resources, Archive) and a generation of apps to host it. The method is deliberately tool-agnostic. People build their second brain in Obsidian, Notion, Tana, Mem, Roam, Apple Notes — whatever holds plain text and lets ideas link to each other. Obsidian became the purist favorite for a specific reason: it is a folder of plain markdown files on your own disk, wired together with backlinks. No server owns your thinking. That is a genuinely great foundation. But there is a hole in almost every second brain, and it is the same hole in all of them. They capture what you type. They lose what you say. --- ## The half of your knowledge that never makes it in Think about where the load-bearing decisions in your week actually happen. The pricing call. The architecture argument. The customer interview where someone said the one sentence that should reshape the roadmap. The 1:1 where a report told you, carefully, that something is wrong. Almost none of that gets typed into your vault. It happens out loud, and then it evaporates. A modern knowledge worker speaks far more words at their computer than they type. Six meetings, a dictated Slack reply between two of them, a voice memo on the way to lunch. The second-brain apps were built for the typing knowledge worker of 2015. The talking knowledge worker of 2026 has a second brain that is, structurally, half-empty. **The gap, in one line** — A second brain that only holds what you typed is a diary of your keyboard, not a record of your work. The decisions live in the conversations — and the conversations live nowhere. The usual patch is a second app. Otter or Fireflies for the meetings, Wispr Flow for the dictation, the second-brain app for everything else. Now you have three knowledge stores that do not know about each other, and "search my second brain" quietly means "search one third of it and hope the rest was important enough to copy over by hand." ## What "one vault" actually means Hyprcore starts from a stubborn premise: every word that comes out of your mouth or your keyboard belongs in the same place. Not synced between three places. The same place. We call it the vault — a real folder tree of markdown pages on your Mac, the same substrate Obsidian fans already trust, with one difference. The meetings and the dictation write directly into it. Concretely, here is what lands in a single vault: - Notes you type — pages, folders, icons, cover images, and a sidebar tree that works the way you think. A real wiki, not a notes drawer. - Meetings you record — when you stop the recorder, the call becomes a vault page: full transcript, an AI summary, action items, speaker labels, and the video under the meta. It is filed, linked, and searchable like any other page, because it is one. - Words you dictate — a global push-to-talk lands transcribed text at your cursor in any app, and anything you dictate into Hyprcore lands in the same tree. Because it is all one store, ⌘K searches across the whole thing in real time — calls, notes, and dictations together. The decision you are hunting for does not care whether you typed it or said it out loud eight days ago. Neither does the search box. > A meeting recorder without a vault is a graveyard for decisions. A vault without your meetings is half a second brain. ## From a personal vault to a company brain The most interesting thing happening in knowledge management right now is the jump from a personal second brain to a shared one — a company brain. Meta has talked publicly about an internal "second brain" rolled out to tens of thousands of employees, with a team-level layer where individual workspaces feed a shared context that agents can read. The pattern repeats everywhere it is tried: the highest-value input is not raw transcripts, it is concise meeting summaries — the decisions, the commitments, the reasoning behind a choice — fed into a store that both people and AI can query. That is exactly the shape a meeting-native vault produces. Every call already resolves to a clean page with a summary, action items, and a link back to the source line of the transcript. You are not building a company brain as a side project; it accretes as a byproduct of recording the meetings you were going to have anyway. Sharing is where most "company brain" stories get nervous, and rightly. So the bar in Hyprcore is end-to-end encryption: sync is opt-in per meeting, the content encrypts with a key derived on your device before it leaves your Mac, and we hold ciphertext, not content. Invite a teammate and the key wraps for them; stop sharing and it goes away. A shared brain should not mean handing your conversations to a vendor in the clear. **Summaries over transcripts** — A 200-line transcript buries the one decision that mattered. The unit of a company brain is the summary — what was decided, who owns it, why. Hyprcore generates that per meeting, and you can edit the prompt so a 1:1 is summarized nothing like a sprint planning. ## A brain your AI tools can actually read A second brain that only a human can open is already behind. The reason the "AI second brain" became a category in the last two years is that a vault of plain text is the perfect thing to point a model at — it is the knowledge base agents search before they answer. The teams getting value out of this are not pasting context into a chat window by hand; their tools reach into the vault directly. Hyprcore exposes the vault over the Model Context Protocol, so an assistant like Claude can list your meetings, read a transcript, pull action items, or look up a person across calls — grounded in what was actually said, with citations, not a hallucinated paraphrase. As of v1.1, connecting it is one click: an Integrations panel wires Hyprcore into Claude Desktop, Codex, Cursor, and other AI tools, with a built-in connection test so you know it is live before your next call. And if you live in the terminal, there is now a read-only "hyprcore" command line — list meetings, read transcripts, and search your vault straight from the shell. And the in-app intelligence is climbing toward the same thing from the other side. The recorder's Ask tab answers questions live during a call; a structured memory graph (people, topics, and commitments pulled out of every meeting) and cross-meeting AI search are both in progress — ask once, get an answer drawn from every meeting you have ever recorded, sources always linking back to the original line. That is the difference between a brain that stores and a brain that recalls. ## Why local-first is non-negotiable for a brain Obsidian people already know this in their bones: a second brain you do not own is a rental. If the company disappears, your thinking should still open in a text editor. So the meetings and dictation hold to the same rule the notes always did. - Dictation runs on-device. Six local speech models — Whisper Small, Medium, Turbo, Large, and Parakeet V3 — run on Apple silicon. Turn off Wi-Fi and it still works. - Meeting audio is captured and transcribed on your Mac, not uploaded to a vendor first. The transcript is written to disk as markdown you can open without us. - The newest engines (NVIDIA Canary) even translate while you talk, so a non-Latin or multilingual meeting still lands as clean, searchable text in the vault. - Anything that leaves the machine — a cloud AI summary, a shared meeting — is a step you opt into, end-to-end encrypted, not the silent default. This is the same instinct that made Obsidian beloved, extended to the things you say. Plain files, on your disk, that you could walk away with tomorrow. ## How to actually structure it If you are coming from a PARA setup in Obsidian or Notion, the mental model carries over cleanly. A few things that work in practice: 1. Let meetings file themselves, then move them. New vaults seed Work, Personal, and Archive folders. Let calls land, and once a week sweep them into the project or area they belong to — the same review habit PARA already asks of you. 2. Write meeting templates per type. A 1:1, a customer interview, and a standup want different summaries. Edit the prompt each template uses once, and every future call of that type gets summarized the way you actually need it. 3. Link out from the summary. The summary is the page that gets reread; the transcript is the archive. Link the decision in a meeting summary to the project page it affects, and your backlinks start mapping how choices flow through your work. 4. Dictate the thought before it is gone. The fastest capture in any second brain is the one that needs zero app-switching. Push-to-talk into the current page beats opening a note to type three sentences you will lose first. 5. Point your AI at the vault, not your memory. When you need "what did we decide about pricing," ask the tool that can read the vault over MCP rather than reconstructing it by hand. Let the brain do recall. ## Where Hyprcore fits next to the tools you know To be clear about lanes: if you have spent two years curating an Obsidian vault with a hundred plugins, Hyprcore is not asking you to throw it away. Obsidian, Notion, and Tana are excellent at the typed half of a second brain. What they do not do is sit on your calls and turn the spoken half into pages without a copy-paste tax. Hyprcore is the tool for the part those apps leave on the floor — the meetings and the dictation — wired into a vault that behaves like the one you already trust: local markdown, a real tree, backlinks, ⌘K everywhere. If you want one place where the words you type and the words you say finally live together, that is the whole idea. If you are a small team and want that place to quietly become a company brain your AI tools can read, that is the next idea. **The premise, restated** — A second brain should hold everything you know — not just everything you bothered to type. Meetings, dictation, and notes in one local-first vault is what that actually takes. There are a lot of words floating around your day, and most of them are spoken. They should all land somewhere you can get them back. Hyprcore is free to download and runs natively on macOS — try recording one meeting and watch it become a page in a vault that already knows everything else you said this week. --- ### How to record meetings on Mac without uploading your audio to the cloud (2026-05-06) Most AI meeting tools ship your audio to a vendor server before you see a word of transcript. Here is what "local-first meeting recording" actually means on macOS, why it matters, and how to set it up in five minutes. URL: https://www.hyprcore.ai/blog/record-meetings-on-mac-without-the-cloud Author: The Hyprcore Team Tags: meetings, privacy, macOS, how-to You hit "join meeting" and a small pop-up appears: "Otter has joined the call." Or Fireflies. Or Read.ai. A few minutes later, your conversation is a row in a vendor database, processed on someone else's GPU, summarized by a model you cannot inspect, and stored under terms most people scroll past. For a lot of meetings, that is fine. For a lot of others — board reviews, hiring debriefs, customer interviews, anything legal-adjacent — it is not. The default of "ship the audio first, ask questions later" makes meeting AI an awkward fit for most of the conversations that actually matter. There is a better default available on the Mac. This guide is about how to use it. --- ## What "local-first" actually means The phrase gets thrown around. Here is the version that matters in practice: - The audio is captured and processed on your Mac, not uploaded to a vendor server. - The transcription model runs on your CPU or GPU, locally. - The transcript is written to your disk, in a format you own (ideally markdown). - Anything that does leave your machine — for example, an AI summary using a cloud model — is something you opt into for that step, not the default. - If you decide to share the meeting with a teammate, that sync is end-to-end encrypted, not just "encrypted in transit." The honest version: most "AI meeting" products fail one or more of these. Some upload first and process later. Some store transcripts on their servers indefinitely. Some technically encrypt sync but hold the keys. The differences are easy to gloss over and matter a lot. **The litmus test** — If you turn off your Wi-Fi mid-meeting and the transcript still appears, the tool is local-first. If a spinner appears instead, it is not. ## Why on-device works on Apple silicon Up until a few years ago, "local speech-to-text" meant slow, inaccurate, or both. Two things changed. First, OpenAI released Whisper as an open model, and the small variants got fast enough to run on a laptop. Second, Apple silicon shipped a Neural Engine and a unified-memory GPU that turned out to be very good at exactly this kind of inference. On a modern Mac, you can run Whisper Large at faster-than-real-time speeds, with accuracy that competes with the best cloud APIs. NVIDIA Parakeet V3 is a more recent option that is sometimes quicker for live captions. The hardware is no longer the bottleneck. The bottleneck is software that bothers to use it. ## What you need to record meetings privately Three pieces: 1. A Mac with Apple silicon (M1 or newer). Intel Macs work, but the speech models will be CPU-bound and slower. 2. System audio capture. macOS does not let apps record other apps by default. You either install a virtual audio driver or use an app that ships its own (Hyprcore does — no kernel extensions, no driver to manage). 3. A speech model that runs on your machine. Whisper variants are the most common. Parakeet V3 is a strong alternative. Avoid anything that requires an API key to transcribe. If you want diarization — the labels that say "Speaker 1 said X, Speaker 2 said Y" — that requires a second model and is harder to do well on-device. It is solvable, but watch for tools that fake it by uploading to the cloud for "advanced features." ## A five-minute setup with Hyprcore This is the way we set it up. (We obviously make Hyprcore, so consider this a worked example with a bias. The general shape applies to any local-first tool.) 1. Install Hyprcore from the download page. Builds are signed and notarized, so Gatekeeper opens them without warnings. 2. On first launch, the onboarding shows a recording-notice screen before model selection. This is the explicit consent step — your meetings will be captured locally; it is up to you to make sure participants know. 3. Pick a speech model. Whisper Turbo is a good default — fast, high-quality, runs comfortably on an M1. If you have an M2 or newer, Whisper Large is stronger. Parakeet V3 is faster for live captions. 4. Grant microphone and screen recording permissions. Screen recording is required because that is how macOS exposes system audio to applications. The captured stream stays on your machine. 5. Open any meeting — Zoom, Google Meet, Teams, FaceTime, anything that plays audio through your Mac — and hit the global shortcut to start recording. The floating recorder appears with a live transcript tab. When the call ends, hit stop. The recorder collapses, post-processing runs locally, and the meeting becomes a wiki page in your sidebar with the full transcript, an AI-generated summary, and action items. The audio file and the markdown both live on your disk. **On the AI summary** — The summary is the one step that uses a model — local, if you have one configured, or a cloud model if you opt in. You can edit the prompt the summary uses per meeting type, or skip the summary entirely and keep just the transcript. ## Sharing without giving up the keys Most teams need to share meetings sometimes — a customer call with a teammate who could not attend, an interview with a hiring panel. The bar here is end-to-end encryption: the vendor (us, in this case, or whoever you pick) cannot read the meeting even if compelled to. In Hyprcore, sync is opt-in per meeting. The audio and transcript encrypt with a key derived on your device before they leave your Mac. We hold ciphertext, not content. If you invite a teammate, the key wraps for them too. If you stop sharing, the wrapped key goes away. The point is not that one product does this and others do not — it is that you should ask the question. "Where does the audio go, who can decrypt it, and how do I revoke access?" If a tool cannot answer that on its security page in three sentences, treat it like it cannot. ## When you should still use a cloud meeting tool Local-first is not a moral position. There are cases where shipping the audio is fine or even better: - Meetings you would happily publish anyway (public webinars, recorded talks). - Calls where the cloud tool is the meeting platform (you are not adding a third party — the audio is already there). - Cases where you genuinely need the vendor's features, like real-time multi-language translation that is currently hard on-device. The point is to make the choice consciously, instead of letting "the default" pick for you. ## A short checklist Before you trust a tool with your meetings, walk through this: 1. Does the audio leave my machine before transcription? (If yes, it is not local-first.) 2. Does the transcript live on my disk in a format I can open without the vendor app? (Markdown is the gold standard.) 3. If the company disappeared tomorrow, would my meeting history be readable? 4. Are AI features opt-in, or are they always-on? 5. Is sync end-to-end encrypted, or just encrypted in transit? 6. Can I see, in the settings, exactly which model is processing my audio? You can answer those questions for any tool, including ours. We tried to make Hyprcore answer them honestly. Whatever you pick, ask. --- If you want to try the workflow above, Hyprcore is free to download and runs natively on macOS and Windows. Pro plans add cloud sync for shared meetings, more local models, and the wiki search across every transcript. Either way, the audio starts and stays on your Mac. --- ### Why we built Hyprcore: one workspace for the words you type, the words you say, and the words you mean to remember (2026-05-04) A wiki, a meeting recorder, and a dictation app — together in one native macOS workspace. Here is the problem that made us combine them, and why that combination matters more than any one feature. URL: https://www.hyprcore.ai/blog/why-we-built-hyprcore Author: The Hyprcore Team Tags: announcement, philosophy, product A modern knowledge worker speaks more words at their computer than they type. They run six meetings a day, dictate a Slack message between two of them, and try to write a doc in the cracks. Then, twenty minutes later, they cannot find the decision that just happened. The tools we use have not caught up to the way we actually work. Notion holds the docs. Otter holds the meetings. Wispr Flow holds the dictations. The browser holds the AI chat. Search across them is a fantasy. The "second brain" people keep promising is, at best, four disconnected first brains. Hyprcore is our attempt at the simple, slightly stubborn idea: every word that comes out of your mouth or your keyboard belongs in the same place. One wiki. Three ways in. --- ## The three-tab problem Watch yourself the next time you finish a meeting. You probably do some version of this: 1. Open the meeting tool to copy the transcript. 2. Open the notes tool to start a new page and paste it in. 3. Open a chat tool to ask an AI to summarize what you just pasted. 4. Paste the summary back into the notes tool. 5. Lose track of which version is canonical six hours later. Each of those tools is fine in isolation. The cost is the seams. Every copy-paste is a chance to lose context, miss an action item, or break a link. The decisions made in your calls — the actually-load-bearing parts of your work — end up scattered across products that do not know about each other. **The premise** — Notes you type, calls you record, and words you dictate should land on the same page tree. Anything less is a filing system that punishes you for talking. ## What we wanted from a single tool We did not start with a feature list. We started with a list of things we refused to give up. - Local-first by default. Dictation should not need an internet connection. Meeting transcripts should live on disk, in plain markdown, in a folder you can open in Finder. - Native to the Mac. Not Electron, not a web wrapper. Apple silicon does GPU-accelerated speech-to-text faster than most cloud services — that should be the floor, not a premium feature. - A real wiki, not a notes app. Pages, folders, icons, cover images, and a tree that works the way you actually think. Linking is non-negotiable. - A command palette that searches everything. Calls, notes, dictations, all of it. ⌘K to anywhere in under 200ms. - AI as a citizen, not a chatbot. Summaries, action items, and answers should live inside your pages and link back to the source line of the transcript. - Encryption when sharing. Meetings sync end-to-end encrypted, or they do not sync. There is no third option. A few of these are easy to say and hard to ship. Local-first speech-to-text is a different engineering problem than calling an API. Speaker diarization on-device is a research project. A command palette that stays under 200ms with thousands of pages takes a real index. We spent more time on the floor of those features than the ceiling. ## How the three pieces fit ### 1. The wiki Every page in Hyprcore is a markdown file with YAML frontmatter, stored on your disk. You can pick an emoji icon and a gradient cover for any page — note, folder, or meeting. The sidebar is a real folder tree with a workspace switcher. ⌘K opens anywhere and searches every title in real time, with recents and base actions. The wiki is the substrate. Everything else writes into it. ### 2. Meetings Meetings record into a floating recorder window — a small pill you can drag, expand into a panel with three tabs (live transcript, notes, ask), and collapse back out of the way. Traffic-light controls. System and microphone capture. Neural acoustic echo cancellation. Speaker diarization. When you stop the meeting, the panel becomes a wiki page with the transcript, an AI summary, action items, and a video embed under the meta — all linked into your tree. You can edit the prompt the AI uses to summarize each meeting type. A 1:1 should not be summarized like a sprint planning. Templates are yours, not ours. ### 3. Dictation A global push-to-talk shortcut works in any app. Hold the key, talk, release. Transcribed text lands at your cursor. Six local speech models — Whisper Small, Medium, Turbo, Large, and Parakeet V3 — run on Apple silicon. No round-trip to a cloud service. Your words do not leave your machine for dictation-only use. Cloud providers are available too, for people who want them. The default is local. ## Why all three, in one place A wiki without dictation is slower than your thinking. A meeting recorder without a wiki is a graveyard for decisions. A dictation tool without a place to land is a transcription factory you have to clean up by hand. Together, they compound. Dictation lands in the wiki. Meetings become wiki pages. ⌘K searches across all of it. The AI you ask "what did we decide about pricing?" answers from your calls, your notes, and your dictations at once, with citations that jump to the source line. **The shape we are after** — A workspace where the act of saying something out loud — to a teammate, to your laptop, to yourself at 2am — is the same act as writing it down. ## What we are not We are not a meeting bot that joins your calls from a Zoom room. We do not store your audio in a vendor cloud you cannot inspect. We are not training models on your meetings — that is in the terms because we mean it. We are also not trying to be a Notion replacement for the entire org. Hyprcore is a workspace for one person and the small team they collaborate with most. If you are running a 200-person knowledge base, you have other tools. If you are running a 200-person knowledge base from inside a 6-person engineering pod, we are probably the right shape. ## Where we go from here The current version of Hyprcore — v0.7.0 — already feels like the product. The floating recorder, the page chrome, the command palette, and the wiki are all in. The roadmap from here is mostly making the seams between them invisible: faster search, more linking patterns, better summaries, MCP exposure to other AI tools so the wiki you are building shows up wherever you happen to be working. If any of that resonates — if you have ever opened your laptop, looked at three different note apps, and quietly given up on finding what you said yesterday — we built this for you. Try it out. Tell us what is missing. There are a lot of words floating around your day. They should all land somewhere. --- ## Comparisons Canonical: https://www.hyprcore.ai/compare Side-by-side comparisons against named competitors. Every competitor claim is cited to that vendor's own pricing page, privacy policy, or documentation, and each page states the date those sources were last checked. Written by the Hyprcore team, who make one of the products being compared — where a competitor is better on a given dimension, these pages say so. ### Hyprcore vs Wispr Flow Canonical: https://www.hyprcore.ai/compare/wispr-flow Competitor: Wispr Flow (https://wisprflow.ai/) — macOS, Windows Claims last verified: 2026-07-15 Verdict: Wispr Flow is the most polished dictation product on the market and it runs on Windows, which Hyprcore never will. Hyprcore now matches it on streaming dictation and adds the thing it has no answer to — Action Mode, where a spoken instruction runs as a command against your selected text instead of being typed out. Pick Wispr Flow for Windows or for cloud accuracy on hard audio; pick Hyprcore for voice that acts, and for keeping it on-device. Feature comparison: - Free tier: Hyprcore — Yes — local dictation + meeting recording; Wispr Flow — Yes — 2,000 words/week on desktop - Opt out of AI model training: Hyprcore — Not applicable — runs on-device; Wispr Flow — Privacy Mode, on every plan - Local on-device transcription: Hyprcore — Yes (Free + Pro); Wispr Flow — No — always cloud - Apple Silicon GPU acceleration: Hyprcore — yes; Wispr Flow — no - macOS native app: Hyprcore — yes; Wispr Flow — yes - Windows: Hyprcore — no; Wispr Flow — yes - Linux: Hyprcore — no; Wispr Flow — no - Works offline (dictation): Hyprcore — Yes — with local model; Wispr Flow — no - System-wide dictation in any app: Hyprcore — yes; Wispr Flow — yes - Streaming dictation (words appear live as you speak): Hyprcore — yes; Wispr Flow — yes - Voice commands — Action Mode (say "Hyper, make this shorter" and it runs): Hyprcore — yes; Wispr Flow — no - Custom wake word for voice commands: Hyprcore — yes; Wispr Flow — no - Custom vocabulary (names, acronyms, jargon): Hyprcore — Yes — biases local + cloud; Wispr Flow — yes - Custom keyboard shortcut: Hyprcore — yes; Wispr Flow — yes - AI cleanup / formatting: Hyprcore — yes; Wispr Flow — yes - Meeting recording (system audio + mic): Hyprcore — yes; Wispr Flow — no - AI meeting summaries & action items: Hyprcore — yes; Wispr Flow — no - Speaker identification: Hyprcore — yes; Wispr Flow — no - Screen + mic recording notes (walkthroughs): Hyprcore — yes; Wispr Flow — no - Web page / YouTube captures into a vault: Hyprcore — yes; Wispr Flow — no - AI wiki / notes / pages: Hyprcore — yes; Wispr Flow — no - Cross-meeting Knowledge Base: Hyprcore — Yes (Pro); Wispr Flow — no - MCP integration with external AI tools: Hyprcore — Yes (Pro); Wispr Flow — no - Cloud STT options (Deepgram, Groq): Hyprcore — Yes (Pro); Wispr Flow — Yes (built-in) Choose Hyprcore if… - You want voice to run commands, not just type — Action Mode has no Wispr Flow equivalent - You want streaming dictation and the option to keep it fully on-device - You handle confidential or regulated work and want on-device transcription - You're already on macOS and want Apple Silicon GPU acceleration - You want one Mac app for dictation, meetings, notes, walkthroughs, and web captures - You record meetings and want them searchable in a Knowledge Base - You want to bring your meeting intelligence into Claude Desktop or other tools via MCP Choose Wispr Flow if… - You're on Windows — Hyprcore is macOS-only and that's the clearest reason to pick Wispr Flow - You want cloud transcription's edge on noisy or heavily accented audio, with no local option to configure - You only need dictation and want the most polished single-purpose product - You don't need voice commands, meeting recording, or a wiki - You're already using Wispr Flow and the workflow fits you Sources: - Wispr Flow pricing and free-tier word limits: https://wisprflow.ai/pricing - Wispr Flow data controls — Privacy Mode and where data is stored: https://wisprflow.ai/data-controls - Wispr Flow privacy — cloud-only transcription: https://wisprflow.ai/privacy - Wispr Flow docs — weekly word limits and resets: https://docs.wisprflow.ai/articles/3715544288-weekly-word-limit-not-resetting-troubleshooting-guide - Wispr Trust Center — SOC 2, ISO 27001, infrastructure: https://trust.wispr.ai - Hyprcore pricing: https://www.hyprcore.ai/pricing ### Hyprcore vs Granola Canonical: https://www.hyprcore.ai/compare/granola Competitor: Granola (https://granola.ai/) — macOS, Windows, iOS, Android Claims last verified: 2026-07-15 Verdict: Granola is the best-designed meeting note-taker on the Mac, and if meeting notes are all you need, its focus is a feature rather than a limitation. Hyprcore is the broader bet — meetings, dictation, and an AI wiki in one app, with a free tier that can transcribe entirely on-device. Pick Granola for the polish; pick Hyprcore if you'd otherwise be paying for three tools, or if the audio can't go to a cloud. Feature comparison: - Free tier: Hyprcore — Yes — local STT, basic recording; Granola — Yes — 30-day note history - Opt out of AI model training: Hyprcore — Not applicable — runs on-device; Granola — One click, but on by default - Stores your meeting audio: Hyprcore — Yes — on your Mac; Granola — No — transcript only - HIPAA compliance: Hyprcore — no; Granola — No — no timeline stated - Local on-device transcription: Hyprcore — Yes (Free + Pro); Granola — no - Apple Silicon GPU acceleration: Hyprcore — yes; Granola — yes - macOS native app: Hyprcore — yes; Granola — yes - Windows: Hyprcore — no; Granola — yes - iOS / Android: Hyprcore — no; Granola — yes - Linux: Hyprcore — no; Granola — no - Records without a bot joining the call: Hyprcore — yes; Granola — yes - Meeting recording (system audio + mic): Hyprcore — yes; Granola — yes - Live transcription: Hyprcore — yes; Granola — yes - AI summaries & action items: Hyprcore — yes; Granola — yes - Live rolling summary (builds during the call, not after): Hyprcore — yes; Granola — yes - Speaker identification: Hyprcore — yes; Granola — yes - Cross-meeting search / Knowledge Base: Hyprcore — Yes (Pro); Granola — yes - Session chat (ask questions about a meeting): Hyprcore — yes; Granola — yes - System-wide dictation (any app): Hyprcore — yes; Granola — no - Voice commands — Action Mode ("Hyper, make this shorter"): Hyprcore — yes; Granola — no - Screen + mic recording notes (walkthroughs with chapters & AI doc): Hyprcore — yes; Granola — no - Web page / YouTube captures into the vault: Hyprcore — yes; Granola — no - Agent that can search, read, and write in your vault: Hyprcore — yes; Granola — no - AI wiki / pages with icons & covers: Hyprcore — yes; Granola — no - Choose between local + cloud STT: Hyprcore — yes; Granola — no - MCP integration with external AI tools: Hyprcore — Yes (Pro); Granola — no - Shareable meeting links: Hyprcore — Yes (Pro); Granola — yes Choose Hyprcore if… - You want one Mac app for meetings, dictation, and notes - You handle confidential meetings and want on-device transcription - You want to choose between local STT and cloud STT per-recording - You also dictate throughout the day in any app - You want meeting intelligence accessible from Claude Desktop via MCP - You want pages with icons and covers as part of the same workspace Choose Granola if… - You only need meeting notes and want the most polished single-purpose product - Granola's opinionated meeting templates fit your workflow - You don't need dictation or a separate notes/wiki tool - Your meetings aren't sensitive enough to require local processing - You're already paying and the workflow fits you Sources: - Granola pricing and model-training FAQ: https://www.granola.ai/pricing - Granola security — no bot, AWS US storage, retention: https://www.granola.ai/security - Granola privacy policy — training opt-out and data location: https://docs.granola.ai/help-center/policies/privacy-policy - Granola docs — subscriptions, billing, and the 30-day free history window: https://docs.granola.ai/help-center/managing-your-account/subscriptions-and-billing - Hyprcore pricing: https://www.hyprcore.ai/pricing ### Hyprcore vs Otter.ai Canonical: https://www.hyprcore.ai/compare/otter-ai Competitor: Otter.ai (https://otter.ai/) — Web, macOS, Windows, iOS, Android Claims last verified: 2026-07-15 Verdict: Otter is the default name in meeting transcription and it goes everywhere a browser or a phone does, which Hyprcore does not. The costs are that everything runs on Otter's servers, the free tier is capped on monthly minutes, and it does meetings and nothing else. Pick Otter for reach and mobile capture; pick Hyprcore on macOS for on-device transcription and for meetings, dictation, and notes in one app. Feature comparison: - Free tier: Hyprcore — Yes — local STT, basic recording; Otter.ai — Yes — 300 min/month, no rollover - Free tier limit per conversation: Hyprcore — No cap; Otter.ai — 30 minutes - Free tier conversation history: Hyprcore — Unlimited — files on your Mac; Otter.ai — 25 most recent, older archived - Opt out of AI model training: Hyprcore — Not applicable — runs on-device; Otter.ai — Enterprise workspaces only - HIPAA compliance: Hyprcore — no; Otter.ai — Enterprise add-on only - Local on-device transcription: Hyprcore — Yes (Free + Pro); Otter.ai — no - Works offline (local-first meetings + dictation): Hyprcore — Yes — with local model; Otter.ai — no - Native macOS app (Apple Silicon): Hyprcore — yes; Otter.ai — Yes — desktop download - Web app: Hyprcore — no; Otter.ai — yes - iOS / Android: Hyprcore — no; Otter.ai — yes - Windows: Hyprcore — no; Otter.ai — yes - Live meeting transcription: Hyprcore — yes; Otter.ai — yes - AI meeting summaries & action items: Hyprcore — yes; Otter.ai — yes - Live rolling summary (builds during the call): Hyprcore — yes; Otter.ai — yes - Speaker identification: Hyprcore — yes; Otter.ai — yes - Cross-meeting search / Knowledge Base: Hyprcore — Yes (Pro); Otter.ai — yes - System-wide dictation in any app: Hyprcore — yes; Otter.ai — no - Streaming dictation (words appear live as you speak): Hyprcore — yes; Otter.ai — no - Voice commands — Action Mode ("Hyper, make this shorter"): Hyprcore — yes; Otter.ai — no - Custom vocabulary (names, acronyms, jargon): Hyprcore — Yes — biases local + cloud; Otter.ai — yes - Screen + mic recording notes (walkthroughs with chapters & AI doc): Hyprcore — yes; Otter.ai — no - Web page / YouTube captures into the vault: Hyprcore — yes; Otter.ai — no - Agent that can search, read, and write in your vault: Hyprcore — yes; Otter.ai — no - AI wiki / pages with icons & covers: Hyprcore — yes; Otter.ai — no - Zoom / Meet / Teams bot integrations: Hyprcore — no; Otter.ai — yes - CRM integrations (Salesforce, HubSpot): Hyprcore — no; Otter.ai — Yes (Business+) - MCP integration with external AI tools: Hyprcore — Yes (Pro); Otter.ai — no - Choose between local + cloud STT: Hyprcore — yes; Otter.ai — no - Per-meeting minute limits on free tier: Hyprcore — no; Otter.ai — yes Choose Hyprcore if… - You're on macOS and want a native Mac app, not a browser tab - You handle confidential meetings and want on-device transcription - You also dictate throughout the day in any app — and want voice commands, not just voice typing - You want a single workspace for meetings, dictation, notes, walkthroughs, and web captures - You want to record a screen-and-mic walkthrough that turns into a searchable note with chapters and an AI-written doc - You want meeting intelligence accessible from Claude Desktop via MCP - You're tired of Otter's monthly minute caps on the free tier Choose Otter.ai if… - You need cross-platform support (web, iOS, Android, Windows) - You record meetings primarily from your phone - You need Zoom / Meet / Teams meeting bot integrations - You need Salesforce or HubSpot CRM integrations on Business+ - You don't need dictation or a wiki - Your meetings aren't sensitive enough to require local processing Sources: - Otter pricing and free-plan limits (prices in USD): https://otter.ai/pricing - Otter privacy policy — training on de-identified data, AWS US storage: https://otter.ai/privacy-policy - Otter privacy and security — SOC 2, HIPAA, encryption: https://otter.ai/privacy-security - Otter help centre — Basic free plan limits: https://help.otter.ai/hc/en-us/articles/360047538094-Conversation-import-and-app-limits-on-the-Basic-free-plan - Otter help centre — HIPAA is Enterprise only: https://help.otter.ai/hc/en-us/articles/33975072019991-HIPAA-Otter-ai - Hyprcore pricing: https://www.hyprcore.ai/pricing ### Hyprcore vs Fathom Canonical: https://www.hyprcore.ai/compare/fathom Competitor: Fathom (https://www.fathom.ai/) — macOS, Windows, Web Claims last verified: 2026-07-15 Verdict: Fathom gives away unlimited recording and transcription and asks nothing for it, which makes it the right default for most people. Hyprcore is the better pick if your meetings can't sit on a US cloud that trains in-house models on de-identified transcripts by default, or if you want dictation and a notes vault in the same app rather than three subscriptions. Feature comparison: - Free tier: Hyprcore — Yes — local STT + meeting recording; Fathom — Yes — unlimited recordings + transcription - Advanced AI summaries on the free plan: Hyprcore — yes; Fathom — Premium and above - Records without a bot joining the call: Hyprcore — yes; Fathom — Yes — desktop capture in Fathom 3.0 - Meeting transcription can run entirely on-device: Hyprcore — Yes (Free + Pro); Fathom — No — processed in the cloud - Where recordings and transcripts are stored: Hyprcore — Your Mac, plus optional encrypted sync; Fathom — United States cloud - Vendor trains its own AI models on your meeting content: Hyprcore — no; Fathom — Yes — de-identified, opt out in settings - Third-party model providers barred from training on your data: Hyprcore — yes; Fathom — yes - Automatic retention / auto-delete policy: Hyprcore — Yours — the files are on your disk; Fathom — Business plan only - Native macOS app (Apple Silicon): Hyprcore — yes; Fathom — yes - Windows: Hyprcore — no; Fathom — yes - Live rolling summary (builds during the call): Hyprcore — yes; Fathom — yes - Speaker identification: Hyprcore — yes; Fathom — yes - AI meeting summaries & action items: Hyprcore — yes; Fathom — yes - Cross-meeting search / Knowledge Base: Hyprcore — Yes (Pro); Fathom — Team plan and above - System-wide dictation in any app: Hyprcore — yes; Fathom — no - Voice commands — Action Mode ("Hyper, make this shorter"): Hyprcore — yes; Fathom — no - Screen + mic recording notes (walkthroughs with chapters & AI doc): Hyprcore — yes; Fathom — no - Web page / YouTube captures into the vault: Hyprcore — yes; Fathom — no - AI wiki / pages with icons & covers: Hyprcore — yes; Fathom — no - Agent that can search, read, and write in your vault: Hyprcore — yes; Fathom — no - MCP integration with external AI tools: Hyprcore — Yes (Pro); Fathom — no - CRM field sync (Salesforce, HubSpot): Hyprcore — no; Fathom — Business plan - Price: Hyprcore — Free · Pro $19.99/mo · Team $39.99/mo (3 seats); Fathom — Free · Premium $20/mo · Team $19/user · Business $34/user Choose Hyprcore if… - Your meetings can't sit on a US cloud — legal, medical, financial, or privileged client work - You want transcription that runs on your Mac, with nothing sent anywhere - You don't want to hunt for a settings toggle to stop a vendor training on your transcripts - You want retention to be a non-question because the files are already on your disk - You also dictate all day and want voice that runs commands, not just types - You want meetings, notes, walkthroughs, and web captures accumulating in one searchable vault - You want your meeting intelligence available in Claude Desktop and other tools over MCP Choose Fathom if… - Anyone on your team is on Windows — Hyprcore is macOS-only and this decides it - You want the most generous free plan in the category, and Fathom's genuinely is - Your meetings are sales calls and you need CRM field sync, Deal View, or coaching scorecards - You want a web app you can open from any machine, not a native Mac app - Cloud storage of meeting content is not a constraint for your organisation - You only need a meeting recorder and won't use dictation or a wiki Sources: - Fathom pricing and plan limits: https://www.fathom.ai/pricing - Fathom privacy policy — AI training and data retention: https://www.fathom.ai/privacy - Fathom help centre — Is Fathom secure?: https://help.fathom.video/en/articles/296512 - Fathom help centre — Retention policies: https://help.fathom.video/en/articles/6057089 - Fathom help centre — Desktop app and bot-free capture: https://help.fathom.video/en/articles/449088 - Hyprcore privacy policy: https://www.hyprcore.ai/privacy ### Hyprcore vs Fireflies.ai Canonical: https://www.hyprcore.ai/compare/fireflies-ai Competitor: Fireflies.ai (https://fireflies.ai/) — Web, macOS, Windows, iOS, Android Claims last verified: 2026-07-15 Verdict: Fireflies wins on reach and integrations, and its Terms of Service bar it from training generative models on your content — a firmer commitment than most competitors make, ours included in writing. Hyprcore is the better pick on macOS if you want the files on your own disk, want named speakers without a bot in the call, or are tired of a 400-minute storage ceiling that never resets. Feature comparison: - Free tier: Hyprcore — Yes — local STT + meeting recording; Fireflies.ai — Yes — unlimited transcription, capped storage - Free plan storage: Hyprcore — Your disk; Fireflies.ai — 400 minutes total — does not reset monthly - Maximum meeting length: Hyprcore — No cap; Fireflies.ai — 2h Free & Pro · 3h Business · 4h Enterprise - Download your transcripts on the free plan: Hyprcore — Yes — they're already files on your Mac; Fireflies.ai — No — paid plans only - Records without a bot joining the call: Hyprcore — yes; Fireflies.ai — Yes — desktop app, or Chrome extension on Meet - Named speaker identification without a bot in the call: Hyprcore — yes; Fireflies.ai — No — labelled Speaker 1, Speaker 2 - Keeps the audio or video file in bot-free mode: Hyprcore — yes; Fireflies.ai — No — transcript and summary only - Bot can be hidden from the participant list when used: Hyprcore — No bot exists; Fireflies.ai — No — always visible, by design - Meeting transcription can run entirely on-device: Hyprcore — Yes (Free + Pro); Fireflies.ai — No — processed in the US - Vendor trains generative AI models on your content: Hyprcore — no; Fireflies.ai — No — prohibited in their Terms of Service - Automatic retention / auto-delete policy: Hyprcore — Yours — the files are on your disk; Fireflies.ai — Enterprise only - HIPAA Business Associate Agreement: Hyprcore — no; Fireflies.ai — Enterprise only - Choose where data is stored: Hyprcore — Your Mac; Fireflies.ai — Enterprise — processing stays in the US - Native macOS app (Apple Silicon): Hyprcore — yes; Fireflies.ai — yes - Windows: Hyprcore — no; Fireflies.ai — yes - iOS / Android: Hyprcore — no; Fireflies.ai — yes - Live transcription during the call: Hyprcore — yes; Fireflies.ai — yes - AI summaries & action items: Hyprcore — yes; Fireflies.ai — yes - Cross-meeting search / Knowledge Base: Hyprcore — Yes (Pro); Fireflies.ai — yes - Third-party app integrations: Hyprcore — MCP into Claude Desktop and similar; Fireflies.ai — 50+ apps, on every plan - System-wide dictation in any app: Hyprcore — yes; Fireflies.ai — no - Voice commands — Action Mode ("Hyper, make this shorter"): Hyprcore — yes; Fireflies.ai — no - Screen + mic recording notes (walkthroughs with chapters & AI doc): Hyprcore — yes; Fireflies.ai — no - Web page / YouTube captures into the vault: Hyprcore — yes; Fireflies.ai — no - AI wiki / pages with icons & covers: Hyprcore — yes; Fireflies.ai — no - Agent that can search, read, and write in your vault: Hyprcore — yes; Fireflies.ai — no - Price: Hyprcore — Free · Pro $19.99/mo · Team $39.99/mo (3 seats); Fireflies.ai — Free · Pro $18/seat · Business $29/seat · Enterprise $39/seat Choose Hyprcore if… - You want the recording and transcript as files on your own Mac, not in a pooled cloud quota - You want named speakers without inviting a bot into the call - You want to keep the audio, not just a transcript, when recording bot-free - A 400-minute lifetime storage cap is not enough and you don't want to pay to fix it - You want to export or delete meetings without a plan gate - You also dictate all day and want voice that runs commands, not just types - You want meetings, notes, walkthroughs, and web captures in one searchable vault Choose Fireflies.ai if… - Your team is on Windows, mobile, or mixed platforms — Hyprcore is macOS-only - You need meetings pushed into Salesforce, HubSpot, a dialer, or an ATS automatically - Your compliance team requires a signed HIPAA BAA — Fireflies Enterprise offers one and Hyprcore does not - You want a contractual no-training commitment in the vendor's Terms, which Fireflies has - You need a workspace-wide auto-delete policy administered centrally - You want a web app reachable from any machine, not a native Mac app Sources: - Fireflies pricing and plan limits: https://fireflies.ai/pricing - Fireflies Terms of Service — AI training restrictions: https://fireflies.ai/terms-of-service - Fireflies privacy policy: https://fireflies.ai/privacy-policy - Fireflies help centre — Meeting storage limits: https://guide.fireflies.ai/articles/4777633252-learn-about-meeting-storage-limits - Fireflies help centre — Recording without a bot: https://guide.fireflies.ai/articles/6666374717-how-to-record-meetings-without-a-bot-on-the-fireflies-desktop-app - Fireflies help centre — How the bot joins your meetings: https://guide.fireflies.ai/articles/9554534786-how-fireflies-joins-and-records-your-meetings-faqs - Fireflies help centre — Auto-delete and retention: https://guide.fireflies.ai/articles/7870593373-learn-about-the-auto-delete-meeting-feature - Fireflies security — HIPAA and compliance: https://fireflies.ai/security - Hyprcore privacy policy: https://www.hyprcore.ai/privacy --- ## Alternatives Guides Canonical: https://www.hyprcore.ai/alternatives ### Best Otter.ai alternatives in 2026 Canonical: https://www.hyprcore.ai/alternatives/otter-ai Claims last verified: 2026-07-15 Disclosure: written by the Hyprcore team, who make the product listed at #1. Otter is the default name in AI meeting transcription, but it's cloud-only, capped on minutes at the free tier, and meeting-only. Here are seven alternatives — ranked, compared, and honest about trade-offs. Ranked: 1. Hyprcore (/) — best for macOS users who want one native app for meetings, dictation, walkthrough recordings, web captures, and an AI wiki — with on-device transcription as an option.. Free; Pro $19.99/mo; Team $39.99/mo (3 seats). · macOS (Apple Silicon and Intel) and Windows x64, both native. · Free tier transcribes on-device with a local speech model. Pro adds optional cloud STT (Deepgram, Groq), switchable per-recording. 2. Granola (https://granola.ai/) — best for macOS users who want the most polished single-purpose meeting note-taker.. Basic $0; Business $14/user/mo; Enterprise $35/user/mo. Monthly billing only — no annual option. · macOS and Windows desktop, plus iPhone and Android. · US AWS VPC, encrypted at rest and in transit. Stores no audio — transcript only. Trains on anonymised data by default, one-click opt-out on every plan, off by default on Enterprise. No HIPAA compliance and no stated timeline. 3. Fathom (https://www.fathom.ai/) — best for Anyone who wants the most generous free plan in the category, and sales teams needing CRM sync.. Free; Premium $20/mo ($16 annually); Team $19/user ($15); Business $34/user ($25). · macOS and Windows desktop apps, plus web. · US cloud, retained until deleted. Trains its own models on de-identified meeting content by default — opt out in settings. Third parties are barred from training. Retention policies are Business-plan only. 4. tldv (https://tldv.io/) — best for Free unlimited recordings across Zoom and Google Meet.. Free tier + paid plans — verify current quotas. · Web, Chrome extension, meeting bots. · Cloud-based. 5. Fireflies.ai (https://fireflies.ai/) — best for Sales and CS teams that need transcripts piped into HubSpot, Salesforce, Slack, and similar.. Free; Pro $18/seat/mo ($10 annually); Business $29 ($19); Enterprise $39, annual only. · Web, macOS, Windows, iOS, Android, Chrome extension. · US cloud (AWS and GCP). Its Terms bar training generative models on your content, on every plan — a firmer commitment than most. HIPAA BAA and auto-delete are Enterprise-only. 6. MacWhisper (https://goodsnooze.gumroad.com/l/macwhisper) — best for macOS users who want to transcribe existing audio files locally on-device.. Free tier + paid Pro. · macOS only. · Fully local — uses Whisper locally on your Mac. 7. Self-hosted Whisper (https://github.com/openai/whisper) — best for Developers who want full control and can build the rest themselves.. Free (compute is your responsibility). · Anywhere you can run Python or whisper.cpp. · Fully local if you run it locally. Sources: - Otter pricing and free-plan limits (prices in USD): https://otter.ai/pricing - Otter help centre — Basic free plan limits: https://help.otter.ai/hc/en-us/articles/360047538094-Conversation-import-and-app-limits-on-the-Basic-free-plan - Otter privacy policy — training on de-identified data: https://otter.ai/privacy-policy - Otter help centre — HIPAA is Enterprise only: https://help.otter.ai/hc/en-us/articles/33975072019991-HIPAA-Otter-ai - Granola pricing and model-training FAQ: https://www.granola.ai/pricing - Granola security — no bot, AWS US storage, retention: https://www.granola.ai/security - Fathom pricing: https://www.fathom.ai/pricing - Fathom privacy policy — AI training and retention: https://www.fathom.ai/privacy - Fireflies pricing: https://fireflies.ai/pricing - Fireflies Terms of Service — AI training restrictions: https://fireflies.ai/terms-of-service ### Best Fathom alternatives in 2026 Canonical: https://www.hyprcore.ai/alternatives/fathom Claims last verified: 2026-07-15 Disclosure: written by the Hyprcore team, who make the product listed at #1. Say the obvious thing first: Fathom gives away more on its free plan than anyone else here, unlimited recordings and unlimited transcription, and no alternative on this page beats that on volume alone. What you're trading for it is US cloud storage kept until you delete it, in-house AI training that runs unless you switch it off, and an intelligence layer that starts at Premium. Here are seven alternatives, ranked and honest about trade-offs. Ranked: 1. Hyprcore (/) — best for macOS users who want one native app for meetings, dictation, walkthrough recordings, web captures, and an AI wiki — with on-device transcription as an option.. Free; Pro $19.99/mo; Team $39.99/mo (3 seats). · macOS (Apple Silicon and Intel) and Windows x64, both native. · Free tier transcribes on-device with a local speech model. Pro adds seven local engines plus optional cloud STT (Deepgram, Groq), switchable per recording. Notes are portable markdown on disk. No bot ever joins a call. 2. Granola (https://granola.ai/) — best for People who want the most polished meeting note-taker and would rather no audio existed at all.. Basic $0; Business $14/user/mo; Enterprise $35/user/mo. Monthly billing only, no annual option. · macOS and Windows desktop, plus iPhone and Android. · US AWS VPC, encrypted at rest and in transit, backed up daily. Stores no audio, transcript only. Trains on anonymised data by default with a one-click opt-out on every plan, and Enterprise is opted out by default. Third parties including OpenAI and Anthropic are barred from training. Not HIPAA compliant, with no stated timeline. 3. Fireflies.ai (https://fireflies.ai/) — best for Teams that need transcripts in HubSpot, Salesforce, and Slack, and want the training question settled in writing.. Free $0; Pro $18/seat/mo ($10 annually); Business $29 ($19); Enterprise $39, annual only. · Web, macOS, Windows, iOS, Android, Chrome extension. · US cloud on AWS and GCP. Its Terms of Service state it "will not use your User Content to train, retrain, fine-tune or otherwise improve any generative artificial intelligence models", on every tier. Zero Data Retention with processing partners. SOC 2 Type II and GDPR. Auto-delete and a HIPAA BAA are Enterprise-only. 4. Otter.ai (https://otter.ai/) — best for People who need meeting notes on a phone, a tablet, and a browser as much as on a laptop.. Basic free; Pro $16.99/user/mo ($8.33 annually); Business $30/user/mo ($19.99 annually); Enterprise on request. USD. · Web, macOS and Windows desktop, iOS, Android, Chrome extension. · AWS S3 in the US, AES-256. Trains proprietary AI on de-identified audio and transcriptions by default; the opt-out is documented only as an Enterprise workspace property, where it is the default, and no self-serve opt-out is documented for Basic, Pro, or Business. Third-party LLMs are barred from training. HIPAA is an Enterprise-only add-on. 5. tldv (https://tldv.io/) — best for Teams whose calls all happen in Zoom or Google Meet and who want a simple bot recorder.. Free tier plus paid plans, historically generous on the free end. · Web, Chrome extension, meeting bots for Zoom and Google Meet. · Cloud-based. We have not independently verified its current data-handling or training terms, so treat the privacy question as open until you read the policy yourself. 6. MacWhisper (https://goodsnooze.gumroad.com/l/macwhisper) — best for macOS users who want to transcribe audio files they already have, entirely on-device.. Free tier plus a paid Pro licence. · macOS only. · Fully local. Whisper runs on your Mac, and nothing is uploaded. 7. Self-hosted Whisper (https://github.com/openai/whisper) — best for Developers who want total control and are willing to build everything around the model.. Free. Compute is your responsibility. · Anywhere you can run Python or whisper.cpp. · Fully local if you run it locally. Nobody's policy applies but your own. Sources: - Fathom pricing (Free, Premium, Team, Business): https://www.fathom.ai/pricing - Fathom privacy policy — in-house AI training, US storage, retention: https://www.fathom.ai/privacy - Fathom help centre — data storage, retention, and deletion: https://help.fathom.video/en/articles/296512 - Fathom help centre — retention policies are Business-plan only: https://help.fathom.video/en/articles/6057089 - Fathom help centre — desktop capture without a bot: https://help.fathom.video/en/articles/449088 - Granola pricing and plans: https://www.granola.ai/pricing - Granola security — no audio stored, AWS US, retention: https://www.granola.ai/security - Granola privacy policy: https://docs.granola.ai/help-center/policies/privacy-policy - Fireflies pricing: https://fireflies.ai/pricing - Fireflies Terms of Service — no generative-model training: https://fireflies.ai/terms-of-service - Fireflies guide — meeting storage limits: https://guide.fireflies.ai/articles/4777633252-learn-about-meeting-storage-limits - Otter pricing (prices in USD): https://otter.ai/pricing - Otter privacy policy — training on de-identified data: https://otter.ai/privacy-policy - Otter help centre — Basic free plan limits: https://help.otter.ai/hc/en-us/articles/360047538094-Conversation-import-and-app-limits-on-the-Basic-free-plan ### Best Fireflies.ai alternatives in 2026 Canonical: https://www.hyprcore.ai/alternatives/fireflies-ai Claims last verified: 2026-07-15 Disclosure: written by the Hyprcore team, who make the product listed at #1. Fireflies is a strong product with an unusually clean stance on AI training, so this page is not a privacy hit piece. People leave over the lifetime storage cap, the paywalled transcript download, and what bot-free recording costs them. Here are seven alternatives, ranked and honest about trade-offs. Ranked: 1. Hyprcore (/) — best for macOS users who want meetings, dictation, walkthrough recordings, web captures, and an AI wiki in one native app — with the files on their own disk.. Free; Pro $19.99/mo; Team $39.99/mo (3 seats, flat). · macOS (Apple Silicon and Intel) and Windows x64, both native. · Free tier transcribes on-device with a local speech model. Pro adds optional cloud STT (Deepgram, Groq), switchable per recording. Notes are portable markdown on your disk. 2. Fathom (https://www.fathom.ai/) — best for Anyone who wants the most generous free plan in this list, and teams that need HIPAA without Enterprise pricing.. Free; Premium $20/mo ($16 annually); Team $19/user ($15); Business $34/user ($25). · macOS and Windows desktop apps, plus web. · All data stored in the United States, retained until actively deleted, and 30 days after account closure. Trains its own in-house models on de-identified Meeting Content by default, with an opt-out in account settings. OpenAI, Anthropic, and Google are explicitly barred from training on it. Retention policies are a Business-plan feature. 3. Granola (https://granola.ai/) — best for People who want the most polished single-purpose meeting note-taker and never want a bot in the call.. Basic $0; Business $14/user/mo; Enterprise $35/user/mo. Monthly billing only — no annual discount. · macOS and Windows desktop, plus iPhone and Android. · US AWS VPC. Stores no audio at all — transcript only. Trains on anonymised data by default with a one-click opt-out on every plan, off by default on Enterprise. States it is not HIPAA compliant and gives no timeline. 4. Otter.ai (https://otter.ai/) — best for Teams that need mobile apps and a browser-first workflow more than they need strong defaults.. Basic free; Pro $16.99/user/mo ($8.33 annually); Business $30/user ($19.99 annually). · Web, macOS, Windows, iOS, Android, Chrome extension. · AWS S3 in the US. Trains on de-identified audio and transcripts by default. The opt-out is documented at Enterprise, where it is the default; below that there is no documented self-serve toggle. HIPAA is an Enterprise-only add-on. 5. tldv (https://tldv.io/) — best for Teams living inside Zoom and Google Meet who want a simple, focused recorder.. Free tier plus paid plans. We have not re-verified tldv's current prices or quotas, so we are not quoting any here. · Web, Chrome extension, and meeting bots for Zoom and Google Meet. · Cloud-based. We have not re-verified tldv's current training, retention, or residency terms, and would rather leave the row blank than guess. 6. MacWhisper (https://goodsnooze.gumroad.com/l/macwhisper) — best for macOS users who want to transcribe audio files they already have, entirely on-device.. Free tier plus a paid Pro licence. · macOS only. · Fully local — runs Whisper on your Mac, so nothing is uploaded. 7. Self-hosted Whisper (https://github.com/openai/whisper) — best for Developers who want total control over the pipeline and can build everything above the model themselves.. Free — compute is your responsibility. · Anywhere you can run Python or whisper.cpp. · Fully local if you run it locally. Nothing leaves your hardware unless you send it somewhere. Sources: - Fireflies pricing and plan limits: https://fireflies.ai/pricing - Fireflies Terms of Service — no generative model training: https://fireflies.ai/terms-of-service - Fireflies privacy policy — no AI training on personal information: https://fireflies.ai/privacy-policy - Fireflies security — HIPAA, SOC 2, and compliance: https://fireflies.ai/security - Fireflies help centre — meeting storage limits: https://guide.fireflies.ai/articles/4777633252-learn-about-meeting-storage-limits - Fireflies help centre — recording without a bot: https://guide.fireflies.ai/articles/6666374717-how-to-record-meetings-without-a-bot-on-the-fireflies-desktop-app - Fireflies help centre — how the bot joins your meetings: https://guide.fireflies.ai/articles/9554534786-how-fireflies-joins-and-records-your-meetings-faqs - Fireflies help centre — auto-delete is Enterprise only: https://guide.fireflies.ai/articles/7870593373-learn-about-the-auto-delete-meeting-feature - Fathom pricing: https://www.fathom.ai/pricing - Fathom privacy policy — AI training and retention: https://www.fathom.ai/privacy - Granola pricing: https://www.granola.ai/pricing - Granola security — no audio stored, US AWS: https://www.granola.ai/security - Otter pricing and free-plan limits: https://otter.ai/pricing - Otter privacy policy — training on de-identified data: https://otter.ai/privacy-policy --- ## Contact Canonical: https://www.hyprcore.ai/contact - Support: support@hyprcore.ai - Privacy inquiries: privacy@hyprcore.ai - Community: Discord at https://discord.gg/UEvrjftB - X / Twitter: https://x.com/hyprcore Typical response time is within 24 hours. --- ## Privacy Summary Canonical: https://www.hyprcore.ai/privacy - On-device dictation processing; text never leaves your machine for dictation-only use. - Meetings sync to encrypted cloud storage only when you opt in or invite teammates. - Transparent third-party integrations for optional cloud models. - You own your meeting data. Every meeting is saved as markdown on your disk. --- ## Terms Summary Canonical: https://www.hyprcore.ai/terms Covers Free and Pro plans, Team sharing, acceptable use, data ownership, and disclaimers. You retain ownership of your content. Hyprcore does not train models on your meeting data.