Changelog

Every release, newest first. Updates are delivered automatically - Dictera checks for them in the background and installs signed, notarized builds.

0.8.5

New

• Pin the languages you actually use. Quick Controls (⌥⌘C) lists every language Dictera handles, and the two or three you work in were scattered somewhere down the alphabet - every switch was a scroll. Click the pin beside a language and it moves to the top, under Pinned. One set covers all three lists: pin Ukrainian once and it is at the top whether you are choosing what you speak, what a dictation is translated into, or what a selection comes back as. Each list shows only the languages it can offer, so a pin your speech model doesn't cover stays out of the spoken list - and stays pinned for when you switch back.

Improvements

• Pinning never moves a row out from under you. The pin fills where you clicked it and the list stays exactly as it was; the new order is there the next time you open it.

0.8.4

Fixes

• "Continue listening" could leave a capture that never heard anything. It is the one button pressed seconds after the Stop it undoes, and starting again while the old session was still unwinding produced a recording that looked alive - the chip said Listening, the timer counted up - while no sound reached the transcriber and the transcript never grew. Bringing the microphone and the system tap up, and taking them down again, now happens one at a time, in the order you pressed things.
• The panel acts on the first click. It is used over the call it is transcribing, so it spends most of a meeting without focus - and macOS swallows the first click into an unfocused window. Mute, Stop and the tabs needed pressing twice. Now the first click is the only one you need.
• A session stays on the day it happened. Reopening Tuesday's call today - to re-read it, ask one more question, or write it up at last - moved it to the top of the list under Today. The list now sorts and groups by when a session began, the time in each row is its start, and a series' open action items are dated by the meeting rather than by the moment somebody summarized it.
• A meeting's length is the meeting's length. Settings ▸ History timed a session from its start to its last activity, so a half-hour call summarized two days later read as two days long. A recording is now measured by what was said in it. The same mistake had old sessions offering themselves up when you attached a recording to today's calendar events; they no longer do.

Improvements

• A Start that has to wait - on a session still unwinding, or on a speech model still downloading - says "Preparing speech recognition…" the moment you press it, instead of leaving the button looking like it missed the press.

0.8.3

New

• Resize the assistant panel - it used to be a fixed rectangle. Drag any edge, drag the dividers between the session list, the conversation and the result, or take the conversation away entirely (⌥⌘T) and give a long write-up the whole window. The size you chose and the spot you dragged the panel to come back next launch, and ⋯ ▸ Reset panel layout puts everything back.
• Every meeting keeps its own setup - shape, output, guests, role and background belong to the session now, not to the app. Last month's town hall keeps its forty people when you reopen it, and setting up a one-off no longer overwrites the standup list you typed once. Settings ▸ AI Assistant holds what new meetings start from; "Save as my defaults" sends a session's setup back there, "Use my defaults" drops it.
• Edit an answer by hand - fix a misheard name, drop a line, add what the model missed, instead of regenerating the nine-tenths that were already right. The model's own words are kept, so Revert still works in a session reopened next week, and anything you changed is disclosed as edited by hand when you copy or save it. Search, the refinement chips and a series' open action items all read your version.
• When a recording runs on - Dictera now offers to stop a recording that has outlived the meeting it belongs to, or one that has been running for hours with nothing scheduled. One minute and one button: Keep recording buys another hour, and the write-up is made either way. Settings ▸ Calendar - and the long-session half works without a calendar connected.
• Paste a picture into a question - ⌘V an image from the clipboard into the panel or the Ask Dictera bar and ask about it: the screenshot you already took with ⌘⇧4, a diagram, a photo. Up to four can travel with one question, and Ask Dictera can now take its own capture too.
• Pick what the Capture button photographs - the whole screen or just the window in front, chosen from the button's own popup and remembered. Whole screens for a call, one window for a coding task.
• Search marks what it matched - the session list, transcripts, answers and templates paint the characters your search found. Matching is Finder's rule, so "resume" finds "résumé" and "OKR" finds "okr".
• Sessions say where their name comes from - a running capture reads "Recording - named when the meeting ends" rather than borrowing a name it is about to lose, and says "Summarizing…" while the name is being written. A recorded session now gets named even if you never asked it for an answer.

Improvements

• Your microphone now starts on in Assist and Meeting Notes - your own voice is in the call from the first word, and one click mutes it. It used to start muted everywhere, which in a notetaker meant your own commitments never reached the action items.
• Stopping a meeting produces whatever that meeting is set to - the notes, full minutes, the transcript or chapters - instead of always the notes. One answer to "what do I get from this meeting" rather than two.
• Dictation now fits itself into Slack, VS Code, Notion, Discord and Teams. Chromium-based apps expose no text to Accessibility until an assistive client asks for it, so the spacing and capitalization at the insertion point had been guessing blind in most chat apps and most editors.
• A screenshot can be removed from a conversation, including one whose picture never came back.
• The panel returns to the spot you last dragged it to, on the display you left it on, and falls back to its default corner when that monitor is gone.
• Binding a meeting from the calendar no longer files the session under a series by itself. Which event this is and which series you file it under are different questions, and a label that appears on its own is one you have to notice before you can undo it - filing stays yours, through the Recurring series field, the row's label picker or the model's suggestion chip. Meeting titles are no longer copied into your preferences file as a side effect.
• All of the above is localized in every supported language.

Fixes

• The countdown before a meeting could appear up to a minute late and then run its full length from there, so recording began after the call had already started. It now fires at exactly the lead time and counts down to the meeting's own start.
• Two meetings that touch: the second one's countdown is no longer lost while the first is still recording, and a Mac that slept through a meeting's end stops the recording instead of counting down to a deadline that passed hours ago.
• Removing a line from a conversation - a screenshot you deleted, or an echo of the other side the app takes back - no longer shifts what the engine has already been shown, which could leave the next question missing a line.
• Switching the answer language or changing a meeting's setup now starts a fresh engine session instead of continuing one that was fixed under the old instructions.
• A cancelled or failed request leaves its lines still owing rather than marking them delivered, so nothing is silently dropped from what the model sees next time.

0.8.2

New

• Calendar - connect your calendar and a recording knows which meeting it belongs to, and who else was in it. Everything is read from the accounts already set up in Calendar - iCloud, Google, Outlook and Exchange - so there is no second sign-in. Read-only: Dictera never writes an event, and nothing leaves your Mac.
• Before a meeting starts - a countdown appears shortly before each meeting and asks whether to record it. Left unanswered it starts recording: a meeting you didn't record is gone, one you didn't want is a click to delete. Set it the other way, or off, per calendar or for one meeting series. Skip the same meeting twice and Dictera stops asking about that one.
• The day's meetings - the session list can show the day's agenda, marking what was recorded, what is still coming and what was missed, with a month grid for reaching any other day.
• Ask Dictera - double-tap right ⇧ for a single field in the middle of the screen. Type the question or dictate it, press Return, and the answer opens in the assistant's Chat tab as a new session. A call the assistant is recording keeps running.
• Recurring series - file a session under the meeting it repeats - Daily, Refinement, Customer interviews - and a year of them stays findable as one set. Filter the session list by series, rename and recolour them, and let a template or a calendar series file new sessions for you.
• What a series still owes - every open action item across every meeting in a series, gathered from notes you already have. No model runs, and it says so when some meetings couldn't be read.
• Label or delete several sessions at once - press and hold a session in the list to start ticking, then file the batch under a series or delete it in one go.
• Capture Front Window - screenshot the window in front instead of the whole screen. A 5K desktop reaches the model with about a tenth of its pixels and its text turns to grey texture; the same budget spent on one window is text the model can read.
• Meeting button makes - pick what the Meeting tab's button produces: the notes, full minutes, the transcript or chapters. Stopping a meeting still always writes the notes.

Improvements

• Two tabs can record the same call at once - start Meeting Notes while Assist is listening and the call is transcribed into both, each in its own session.
• Open another session while an answer is generating and the answer still lands in the session that asked for it. The session you moved to says an answer is finishing elsewhere rather than showing a spinner that isn't its own.
• Sessions can be renamed in place in the list, without opening them.
• Quick Controls say which language each feature will use, and the recording indicator names the language it is transcribing in.
• Attendee names read from your calendar stay on this Mac unless you allow them to reach a remote engine. Email addresses are never stored at all.
• Screenshots come back with a reopened session at the size they were taken, and the viewer shows that size.
• "Claude Code" is now "Claude Code CLI", so the engine that runs on your Mac isn't mistaken for a Claude model reached over an API.
• All of the above is localized in every supported language.

Fixes

• Panel tooltips that pair a name with its shortcut appeared in English whatever the interface language was.
• The Claude Code CLI engine no longer collects a process's output on a lower-priority thread than the one waiting for it, which macOS reports as a hang risk.

0.8.1

New

• Quick Controls - one shortcut opens a compact panel in the middle of the screen for the things you change mid-flow: switch the speech model, change the spoken language, toggle cleanup. No trip to Settings.
• Screenshot viewer - open any screenshot you've sent to the assistant at full size, zoom and pan it, fit it to the window or view it actual-size, then copy or save it.
• Screenshots are kept with the session - reopening a saved conversation brings its pictures back. Storage is capped and the oldest are cleared automatically.
• Model updates - when a newer version of a model you have installed is published, Dictera offers it, verifies it before installing, and reclaims the space of the one it replaces.
• Downloads resume - an interrupted model download picks up where it stopped instead of starting over.
• Connectors, in detail - connector tools are discovered in the background, listed per server, and granted individually. Tools that can change things rather than only read them are marked as such.

Improvements

• Dictated text now fits its surroundings better - Dictera looks at what's immediately before the caret and joins on correctly instead of doubling a space or a full stop.
• Correcting selected text is context-aware: an unchanged result is left alone rather than pasted over itself, and switching apps mid-correction is handled cleanly.
• The translation indicator names the language it's translating into.
• The AI Assistant now needs a subscription once your trial ends, like the rest of the app. During the 14-day trial everything is unlimited. Saved sessions stay readable and exportable either way.

Removed

• Qwen3-ASR is no longer available - the framework Dictera uses for on-device speech removed the engine, so no version of the app can both keep it and take any further update. If you had it selected you'll be moved to Apple Speech Analyzer and offered its disk space back.

0.7.1

New

• Meeting Notes - a second mode for the AI Assistant: it listens to the call and writes it up. Summary, decisions and action items when you press Stop, or any time with the summarize button. The same session can also be produced as Full minutes, a cleaned-up Transcript in the language it was spoken, or Chapters - a timeline of what was discussed when.
• The notes take the shape of the session - a lecture has no decisions and no owners; a meeting does. Left on automatic, each session is read for what it is - meeting, lecture or talk, one-on-one, interview, sales call - and the notes say which shape they used, so a wrong guess is one change away.
• Meeting Setup - name the other participants, who you are, and any background, and an action item can say who owns it. The setup is kept between meetings.
• Chat - a third mode: a plain typed conversation with the model, with no audio captured at all.
• Session list - a sidebar of your conversations for the current mode, searchable and renameable. Reopening one brings its mode back, and the session that is still recording is marked as such.
• One-press rewrites - an answer can be made Shorter or Simpler, or given More detail or an Example. Notes can be cut down to Action items, reorganized By topic, or turned into a Recap email, Key terms or a Study guide.
• Answer and Listen hotkeys - ⌥⌘A answers the last question and ⌥⌘L starts or stops listening, from any app, so you never have to click into the panel - the thing most likely to give the assistant away on camera. In Meeting Notes, Answer writes the notes instead.
• Every panel shortcut is editable, grouped as Actions, Navigation and Display, with Restore Panel Defaults to undo your changes.
• Cancel speaker echo - the microphone pauses while the other side is coming out of your speakers, so they aren't transcribed a second time. On by default, and it arms itself only when you're on loudspeakers.
• Claude Code as an engine - the assistant can talk about your actual project: Claude reads the folder you choose before answering. Access is read-only - it can read and search the folder, never change it. Tick exactly which connector tools it may use, and run a Health Check to see what the folder can reach.
• A model and a language per mode - Assist, Meeting Notes and Chat each pick their own engine, switchable from the panel's toolbar too.
• Compact mode - a smaller panel for when it has to share the screen with a call.
• Save as Markdown… for notes, minutes, transcript and chapters, plus copy buttons for each of them and for a single answer.
• Context Window per remote model - Test Connection fills it in when the server reports it. Getting it right is what lets a long meeting be summarized in as few passes as possible.

Improvements

• A long meeting no longer stops at the model's limit - the transcript is read in passes, the panel says which pass it's on, and the result says when it covers only the most recent stretch.
• A conversation that outgrows the model's context window now says so, and points at where to set a larger one.
• Your microphone starts muted in every mode. Turn Mic on when your own voice should be part of the notes; the panel shows which state you're in.
• The panel says what the engine is doing while you wait - reading a file, searching the project, fetching a page, running a command.
• Screen-sharing visibility is now a button in the panel itself: one click to hide the window from capture or bring it back.
• Answer text can be made bigger or smaller and reset again.
• Pick which output device the other participants are captured from - the one you actually hear the call on (e.g. your AirPods), or System Default.
• All of the above is localized in every supported language.

Fixes

• Dictation into Dictera's own text fields - the assistant's chat box, Settings - now inserts the text there instead of sending it to whatever app sits behind the panel.

0.6.4

• An Open Settings global hotkey.
• A reworked onboarding flow, with live "try it" fields that confirm each step worked.
• Interface and localization updates throughout.

0.6.2

• Edit Selection - select text in any app, hold a key, and say the change you want. The rewrite lands in place, one undo away.
• History gained searchable Assistant Sessions.

0.6.1

• The first AI Assistant: a floating panel with a global hotkey, system audio capture, and reusable session templates.

0.5.1

• Correct Selection and Translate Selection - fix or translate text you've already written, in any app.

Earlier

• 0.4.x - bring your own AI: connect Ollama, LM Studio, or any OpenAI-compatible server, per feature.
• 0.3.x - additional speech models, dictation reliability work, and the first tutorials.

Something missing or broken? info@dictera.app.