Changelog
Follow up on the latest improvements and updates.
RSS
An early beta of Hedy's AI running fully on-device on your phone or tablet.
- Requires iPhone 15 Pro or newer (any iPhone 16 or 17 model qualifies), or an M-series iPad
- Requires iOS or iPadOS 17 or later
- Your conversation data stays on your device for anything a local model handles
- Choose from three compact models tuned for mobile: Qwen 3.5 2B (1.2 GB), Qwen 3.5 4B (2.6 GB), or Gemma 4 E2B (2.9 GB)
- Quality is solid for short summaries and notes; expect slower responses, higher battery use, and occasional instability compared to cloud AI
This is brand-new territory for mobile devices. When you turn it on, Hedy will offer to disable Automatic Suggestions to keep your phone responsive. Try it out from Settings, Speech & AI, and let us know how it feels.
The Parakeet on-device speech engine, already available on Apple Silicon Macs, now works on iPhone too.
- Captures meetings with speaker labels (diarization) baked in, so you see who said what without any setup
- Noticeably better accuracy on accented English and 25+ European languages
- Runs entirely on your device, no network connection required, no audio leaves your phone
- Roughly 2.5 GB one-time model download
- Available on iPhone 15 Pro and newer running iOS 17 or later
- Marked Experimental in Settings, Transcription. Best for English and major European languages; if your transcript comes out in the wrong language, switch back to Whisper
new
iOS
Android
External Microphones for Live Sessions (v3.2)
Use any USB, Bluetooth, or wired microphone for your live sessions on iPhone, iPad, and Android.
- Pick the right mic for the room from a clean selector in Meeting Settings
- On iPhone and iPad, you can switch microphones live, mid-session, without losing your place
- On Android, your selection is applied the next time you start a session
- Your choice is remembered across sessions; if a saved mic isn't connected, Hedy falls back to the system default
- Particularly useful for podcasts, conference rooms, and noisy environments where a clip-on or headset mic makes a real difference
new
Windows
Local AI on Windows (v3.2)
Run Hedy's AI fully on your Windows PC.
- Powered by llama.cpp with Vulkan GPU acceleration and hybrid GPU plus CPU offload, so models that exceed your VRAM still run, just slower
- A live "Memory available for AI" indicator shows how each model fits your hardware before you download
- Choose from compact 2 GB models for smaller systems up to 20 GB+ models for high-end GPUs
- Models that need to fall back to CPU are clearly marked with a "+ Slow" badge
- Your conversation data stays on your machine for anything a local model handles, no round-trip to the cloud
Currently in active development, expect quality and model selection to keep improving.
new
PRO
macOS
iOS
Android
Web
Windows
Webhook Signing Secrets (v3.2)
Cryptographically verify that webhook calls really come from Hedy.
- After you create a webhook, your unique signing secret is shown one time in a confirmation dialog
- Hedy signs every payload with HMAC-SHA256 and sends the signature in the X-Hedy-Signature header
- Verify it on your server to reject spoofed or tampered requests
- The secret cannot be retrieved later, so save it the first time you see it. To rotate, delete and recreate the webhook
- Available on the Pro plan, up to 10 webhooks per account
Find webhooks under Settings, API Access, Manage Webhooks.
Both export tools now work on web.hedy.ai with full parity to the desktop and mobile apps.
- Quick Export: one click to back up all your sessions, topics, and highlights as a single JSON file
- Advanced Export: choose Markdown, CSV, or JSON, scoped to a specific topic, all sessions, or all highlights
- For Markdown and JSON, pick a single combined file or a ZIP with separate files per session
- Toggle exactly what to include: transcript, summary, meeting minutes, highlights, action items, AI insights
- Both exports live under Automation, Backup & Transfer
improved
macOS
iOS
Windows
Local AI Model Management (v3.2)
Manage your downloaded local AI models from Settings, Speech & AI.
- Each model card shows the parameter count, file size, memory required, and a quality rating from one to three stars
- A fit indicator tells you up front whether a model will run comfortably on your machine
- Delete any downloaded model you don't need to free up disk space; re-download on demand later
- Tooltips explain what to expect from each quality tier, from "good for basic summaries" up to "close to our cloud AI"
fixed
macOS
iOS
Android
Windows
Recover Sessions Stuck After Post-Processing (v3.2)
A new recovery card now appears on the Details tab when a session ended but didn't finish processing, for example after a crash, network drop, or LLM timeout.
- Replaces the old confusing layout of empty Summary, To-Dos, and Detailed Notes cards
- Tap "Complete processing" to re-run the full pipeline: title, summary, to-dos, and detailed notes if your settings include them
- Hidden automatically when Cloud AI is turned off, so you don't see a button you can't use
Run Hedy's AI analysis entirely on your Mac.
What it does:
- Performs summaries, suggestions, and insights directly on your Mac.
- Keeps your transcripts on-device — nothing leaves your computer for analysis.
- Works offline once the model is downloaded.
How to enable it:
- Open Settings → Speech & AI.
- Scroll to the Local AI section and turn on Local AI Processing.
- Choose a model tuned to your Mac's hardware.
Good to know:
- Beta feature: quality and speed will keep improving.
- Requires an Apple Silicon Mac with enough free memory for the model you pick.
- You can switch back to cloud analysis anytime.
new
macOS
iOS
Android
Windows
OMI Wearable Integration (v3.2)
Hedy now works with the OMI wearable. Pair your device once and the transcript streams straight into your Hedy session.
- Guided pairing flow walks you through setup.
- Works across iPhone, iPad, Mac, and Windows.
Load More
→