Hedy AI
Roadmap
Feedback
Changelog

Changelog

Follow up on the latest improvements and updates.

RSS

new

PRO

macOS

iOS

Android

Web

Windows

Public Links for Any Session (v3.10)

Publish a session as a web page that opens for anyone you send it to. No Hedy account, no sign-in, nothing to install.
What you choose
  • Which parts appear: the summary, your detailed notes, to-dos, session notes, the transcript, and your chat with Hedy. Pick any combination.
  • How long it lasts: 7, 30 or 90 days, or no expiry at all.
  • You can revoke the link at any time.
Where to find it
  • Open a session, tap
    Share
    , then
    Public link
    .
  • Your links, with their expiry and view count, are listed under
    Settings
    then
    Account
    then
    Sharing
    .
What the reader sees
  • A page styled like Hedy, with each section collapsible so a long transcript does not bury the rest.
  • The speaker names you set, including any speakers you merged.
  • Your name, as the person who shared it.
Sharing now names each option for what it does
  • Invite link
    is single use, expires in seven days, and asks the recipient to sign in.
  • Public link
    opens for anyone holding it.
  • Copy bookmark
    grants no access at all. It only opens for someone who already has it.
Public links need Hedy Pro and cover sessions, not topics. The choices you make are fixed when the link is created, so to change them, revoke it and make a new one. Anyone who already opened a link may have kept a copy, which revoking cannot undo.

new

macOS

iOS

Android

Web

Windows

Merge Speakers Into One (v3.10)

Speaker Names in Your Transcripts arrived in 3.9. Now, when speaker detection hears one person as two, you can join them.
  • Open the transcript, tap a speaker chip, and choose
    Merge into...
    to fold that speaker into another named speaker.
  • They share one name, one colour, and count as one person in your summary and detailed notes. Renaming both halves the same thing never did that.
  • The merge follows the session into copies, downloads, shares, public links and your other devices.
  • Your stored transcript is not rewritten. The merge is applied when the transcript is read.
Worth knowing before you use it: the speaker you merge into must already have a name, and a merge cannot be undone from the app. Available on sessions captured with the Nemotron speech engine, which is what produces speaker-separated transcripts.

improved

iOS

Nemotron on More iPhones and iPads (v3.10)

The Nemotron on-device speech engine, which gives you speaker-separated transcripts, now runs on far more devices.
  • Every iPhone supported by iOS 17, from the iPhone XS and XR onward, instead of starting at the iPhone 12.
  • Every iPad with a Neural Engine, including the 2018 and 2020 iPad Pro.
  • Select it under
    Settings
    then
    Speech & AI
    then
    Speech Recognition Options
    .
If a device cannot prepare the models, Hedy falls back to Whisper by itself and tries again on a later version, so you are never left without capture.

improved

PRO

macOS

iOS

Android

Web

Windows

To-Dos Stay in Step Across Devices (v3.10)

Building on Consolidated Topic To-Dos in 3.9, ticking off an action item now agrees everywhere it appears.
  • Completion matches between the session it came from, the Topic
    To-Dos
    tab, your other devices, and the API.
  • Completion is stored per item instead of by rewriting the whole list, so one device can no longer undo another's change.
  • New for Pro:
    PATCH /v1/sessions/:sessionId/todos/:todoId
    accepts
    completed
    and
    dueDate
    , and MCP gains
    UpdateToDoCompletion
    and
    UpdateToDoDueDate
    , so you can tick items off or re-date them from Claude.
  • Due dates are free text, so "next Friday" and "2026-08-22" both work. An empty string clears one.
  • Editing the item in the app afterwards takes authority back from a due date set through the API.

improved

macOS

iOS

Android

Web

Windows

Clearer Cloud AI Analysis Control (v3.10)

The privacy control is now named after the feature it governs, so the switch position matches what it does.
  • Cloud AI Analysis
    on means Hedy writes your summaries, highlights, meeting notes and chat in the cloud. Off means it does not.
  • With it off, Hedy names what still happens on your device rather than leaving you to work it out.
  • The in-session banner is now reachable from every tab, and its Settings link goes where it should.
  • Find it under
    Settings
    then
    Privacy & System
    .
Your existing setting is unchanged. This is about the label and the switch, not the value behind it.

improved

macOS

iOS

Windows

Steadier On-Device AI (v3.10)

Local AI now runs each model with the sampling settings its makers publish, and constrains its output to the shape each feature needs.
  • Recaps, speaker identification, suggestion analysis and topic overviews can no longer come back structurally valid but empty.
  • Qwen models stop repeating themselves: every repetition measure improved in our own recap and detailed-notes testing, and the loops we could reproduce are gone.
  • If a generation does stall, Hedy now tells you whether that one request stalled or Local AI is unavailable until you restart, rather than going quiet for the rest of the session.
  • Hedy still never falls back to a cloud provider here. Choosing on-device means staying on-device.
  • Under
    Settings
    then
    Speech & AI
    then
    Local AI
    .

new

macOS

Windows

Qwen 3.8 27B for Local AI (v3.10)

A new 27B on-device model joins the Local AI catalog, at a size that fits comfortably where the model it succeeds was a tight squeeze.
  • Qwen 3.8 27B ships at Unsloth's Dynamic v3.0 quant, around 10.93 GB, against roughly 18 GB for Qwen 3.6 27B.
  • On a 24 GB Mac that is the difference between a tight fit and a comfortable one.
  • Download it under
    Settings
    then
    Speech & AI
    then
    Local AI
    .
  • Qwen 3.6 27B is archived rather than removed. If you already downloaded it, it keeps working and you can still delete it to free space.

improved

PRO

Faster MCP Connections and Bounded Session Listings (v3.10)

Connecting an MCP client to Hedy is about a second quicker, and the Zapier session listing is now bounded.
  • MCP OAuth discovery advertises the scopes the server actually supports,
    mcp:read
    and
    mcp:write
    , instead of reading every registered client on each request. No client's granted permissions change.
  • Worth checking if you built on it:
    GET /v1/sessions?format=zapier
    with no
    limit
    or
    after
    used to return every session in one response. It now returns at most 100, newest first, and that response format carries no
    hasMore
    flag. If your automation relied on one unbounded call, page it with
    limit
    and
    after
    .

improved

macOS

Windows

A New Call Can Replace a Paused Session (v3.10)

Automatic meeting detection now handles the case where a session is already paused when your next call starts.
  • Hedy offers
    End previous and start new
    , so two back-to-back calls stay two separate sessions with their own transcripts.
  • Accepting finishes the paused session properly first, then starts the new one.
  • Turn detection on under
    Settings
    then
    Automation & Data
    then
    Meeting app detection
    then
    Auto-detect video calls
    .
  • Calendar-only prompts stay quiet while a session is paused.

improved

macOS

iOS

Android

Windows

Suggestions Run Through Long Sessions (v3.10)

Increased reliability of live suggestions in longer sessions.
  • Hedy now measures the summarised context it actually sends for analysis rather than the full raw transcript, so the cycle keeps going as a conversation grows.
  • Needs automatic suggestions switched on and an available AI provider.
Load More
→