Hey, I'm Adil, founder of Hyprcore, writing from San Francisco.

We've heard from a lot of you on Windows machines who were stuck with a version that didn't feel finished.

As of this week, that's fixed.

Hyprcore v1.11 ships full feature parity on Windows 10 and 11.

Hold Ctrl+Space anywhere and dictate straight into whatever you're typing into, no app switching required.

Meetings record both sides of the call, your mic and whatever's coming out of your speakers, kept separate and mixed cleanly, with no extra audio driver to install.

Screen capture runs alongside the audio, and Obsidian users can wire up their vault with one click.

We also rebuilt the setup flow. Instead of a progress bar and a hope, it ends by proving the app actually works before you start using it.

Same privacy-first architecture as always: everything processes on your own machine, nothing gets shipped off to a server we control.

If you've been waiting for Windows to catch up to Mac, it just did.

Hey Sumo-lings! 👋

A lot of you told us Hyprcore was great at remembering your calls, but you wanted it to actually do something with everything it captured, not just store it. We heard you. Our v1.10 release, out August 14, adds three new capabilities: routines, connectors, and webhooks. Here's what's new and why we built it the way we did.

🧭 It works while you're not there

The most requested feature was some form of automation, so we built routines: describe a job in plain English, tell it when to run, and it happens on its own, before a meeting, after one, or on a schedule, laptop shut or not. No canvas, no nodes to wire up. One of our shipping routines, quoted exactly as it runs:

When: 30 minutes before a meeting with external guests Do: Look up the event, then brief me: who is attending and their company, the last email threads with them, and what my notes say we discussed or promised last time. Keep it under 200 words, and write it into the meeting's Prep section. Can reach: your mail (read), your meetings (read)

We shipped 10 starter routines to get you going, including meeting prep, a morning briefing, a Friday weekly review, an action item sweep, and follow-up drafts.

🔒 You told us you didn't want a black box

The biggest hesitation around letting AI act on your behalf was trust. So permissions are two separate settings, not one slider: machine access (can it touch files and run commands on your Mac?) and state change (can it change anything, anywhere?). Every new routine starts read-only and sandboxed. Autonomy is something you turn on, not something you have to turn off.

📎 Connectors, so it can reach your other tools

Gmail, Outlook mail, Notion, and web search are connected now. Every connector tells you what it can reach in plain language (read your mail, draft replies, send mail) instead of a scope string you have to decode yourself. And any limit you set gets enforced on the request itself before it leaves your machine, so it's a real boundary, not just a setting you hope holds.

🔌 Webhooks, because Hyprcore doesn't need to be your whole stack

If you already run automations elsewhere, a routine can post straight to your own endpoint instead of running the AI at all. Signed, retried automatically, and logged down to the byte. Plain HTTP is allowed on localhost too, so a self-hosted n8n never needs a public relay.

💸 On usage: we heard you there too

Connector calls draw from your monthly allowance, same as before, but we added real ways to run routines without spending it down. Routines that only read your own meetings and notes are unmetered, and you can bring your own API key or run everything through your own Claude Code subscription instead.

Also in this release: Quick Ask, a keyboard shortcut you can summon from anywhere to ask your own meetings and notes a question on the spot.

None of this changes how your data is handled. Everything is still local-first, still yours, and every routine lives as a plain markdown file in your vault that you can open, edit, or delete whenever you want.

Full release notes: hyprcore.ai/updates

Thanks for pushing us here. Keep the feedback coming.

— Adil & the Hyprcore team

Hi, Sumo community! 👋

Since launch 2 weeks ago we've shipped nine releases. Here's what actually changed about using it.

🌍 Your language, your words

Pick the language you're speaking and Hyprcore keeps it that way — or translate to English when that's what you need. And your custom vocabulary is now one shared list across dictation and meetings: teach it your teammates' names and your jargon once, and they come out right everywhere.

📅 Your calendar, inside the meeting

Connect Google or Microsoft Calendar and your upcoming meetings, attendees, and call links show up in Hyprcore — and those attendees feed straight into naming the voices afterwards.

📥 In and out

Import audio, documents, and email exports in batches, and they become searchable context sitting alongside your meetings — worth a look if you're migrating with a back catalog of recordings. Going the other way, download any meeting as Markdown, plain text, or Word. Nothing's locked in here.

🗣️ Transcripts that know who's talking

Most of our engineering went here. Three things that used to happen, and now don't:

- Your own speakers echoing back as a second person repeating what you just heard

- A remote participant renamed to *you* because your mic caught the room

- Long silences coming back as sentences nobody said

You can also review and rename every speaker after a call.

🧠 Twelve new models

DeepSeek V4 Flash is the new default for summaries, chat, and dictation clean-up — it costs a good deal less to run than what it replaced, so your credits go further on the same meetings. Grok STT is the new default for meeting transcription, and Qwen3 ASR Flash is there if you work in Chinese.

Ten more are in Settings → Models.

Full release notes: https://www.hyprcore.ai/updates

Thanks for being part of the Hyprcore family. 🙏

A few of you have asked how tokens actually get used, so here's the straight answer.

Tokens are spent whenever the AI does work for you — transcribing recordings, writing summaries, pulling out action items, chatting with the AI, asking questions, regenerating a summary, or searching your notes. Recording itself is only part of it; the more you lean on the AI afterward, the more you use.

For scale: one hour of recording plus its summary runs around 50,000 tokens. So 500,000 tokens covers roughly 10 hours of light use, and 4,000,000 covers about 70-80 hours.

Tokens aren't your only option, though. You can run transcription and summaries locally on your own machine for zero tokens — fully private, works offline. The tradeoff is that local processing depends on your hardware: a newer Mac handles it comfortably, while a lighter setup may be slower or need smaller models. Cloud tokens are where you get the top-tier models and more speed. Most people mix both — local for everyday recording, cloud when they want the best quality.

One more note on accuracy: our multi-speaker detection isn't perfect yet, especially on busy calls with a lot of crosstalk, since diarization runs locally on your machine. It's an area we're actively improving, and you can always relabel a speaker to correct it in the meantime.

Thanks for the questions — keep them coming.

—The Hyprcore team

Hey AppSumo fam! 🎉

We built Hyprcore because we were tired of juggling five different tools just to remember what happened in a call.

Hyprcore is a lightning-fast desktop workspace that records meetings, transcribes every word locally, and stitches them into a searchable company brain—no bots, no clouds, no creepy minute meters. 🔒

Because everything happens on your own machine, you get unlimited transcription and iron-clad privacy while we pick up the tab for the heavier AI summarization you actually care about.

That efficiency is what lets us offer lifetime access where cloud tools simply can’t. 🚀

We’re beyond excited to drop Hyprcore on AppSumo and see what this community will build when every conversation is instantly at your fingertips.

Let’s turn talk into leverage together! 💪