Dicta

Dicta 更新日志

App 每天检查一次更新,安装前会显示这些说明。

所有 App 的更新日志
更新说明语言English简体中文
  1. 0.1.0-beta.9Beta

    Added

    • A feedback window: Help → Send Feedback now opens a window of its own. Choose a type (Problem, Suggestion, Question or Other), describe what happened, add screenshots or a screen recording (drop them on the window) and preview them there — the video plays in the window — and review the diagnostics before you send. Settings → Feedback lists what you have sent and the replies.
    • “Report this problem…” where something fails — when the engine stops, a transcription fails, a file in a batch fails or is not exported, a model does not download, or an import refuses files. It opens the feedback window with the error, where and when it happened and your last few steps filled in, plus the related log lines with personal details removed; you can read the report and leave it out.

    Changed

    • Settings → Feedback no longer has the form: it lists your feedback, with a button to open the feedback window.
  2. 0.1.0-beta.8Beta

    Added

    • A 15-day free trial: Dicta now runs for 15 days on each Mac from its first launch, without a code or an account. Settings → Licence shows the days left and the day the trial ends; in its last 3 days a line in the sidebar says so (click it to buy a licence). When the trial is over, Dicta asks for a licence code; your recordings, transcripts and settings are kept. A licence code always replaces the trial. Early-access codes already given keep working until they end.
    • Send feedback from inside Dicta (Settings → Feedback, or Help → Send Feedback): describe what happened, add up to 3 screenshots or a screen recording, and choose whether to include diagnostics (keys, passwords, names, email and IP addresses removed; “See what is sent” shows them first). We reply by email, and replies also appear under “Your feedback”, where you can answer.
    • Agents: more to give a transcription. An agent that starts a transcription can add names and terms to expect, one of your vocabulary sets, a summary template, and show notes (kept as the transcript's note, which you can edit). Agents can also list your templates and people.
    • Agents: people and vocabulary sets (Settings → Agents → “Allow agents to manage people and vocabulary sets”; off by default, and Dicta asks before each program's first change). An agent can add people (a podcast's guests, with their role, a short bio and links) and vocabulary sets, and suggest who a transcript's speakers are; you confirm each suggestion. Agents never delete anything or learn a voice, and change only what agents added. People show who added them; you can edit or delete them on the People page.
    • Download links that need a sign-in with your browser's: when a video needs you signed in, or YouTube blocks your network (“confirm you're not a bot”), Dicta finds the browser on this Mac that is signed in to that website (Chrome, Arc, Edge, Brave, Firefox, Chromium, Vivaldi or Opera; your default browser first, else the one used last; it looks only at the names of the website's cookies, which asks for no permission) and asks “Sign-in needed to download: continue with Arc's sign-in?” [Continue]. You never pick a browser. With none signed in, it says to sign in in your browser and click Try again, which looks again; “Other ways” takes a cookies.txt file. Dicta reads the browser's sign-in itself, only after you continue, once per download, and only that website's cookies, into a file only you can read, deleted when the download ends; it saves nothing. After such a download Dicta asks once whether to do it by itself from now on: Settings → Import & export → “Use your browser's sign-in when a link needs one” (off by default) then goes on without asking. For Chrome, Arc and the other Chromium browsers macOS asks whether Dicta may use the browser's “Safe Storage” (each download, or never again with “Always Allow”, which lets only Dicta read it); Firefox needs no permission. When the browser has no sign-in to the website, or macOS did not allow reading it, Dicta says so.
    • Settings → Phone: open or scan a QR code to get the iPhone app “Dicta Recorder” (TestFlight public beta).
    • Try again from the ⋯ menu: a failed transcription's ⋯ menu (in the library and on its page) has “Try again”, and a stopped one's “Resume” or “Continue”, near the top.

    Changed

    • The activation page's “Request early access” is now “Buy a licence”, which opens Dicta's page on actnow.dev.

    Fixed

    • YouTube's bot check (“Sign in to confirm you're not a bot”) said the video needs a sign-in; it now says YouTube is blocking the network, with what to try.
    • Sign-in cookies were used for a link's download but not when its details were read first, so a video that needs a sign-in still failed.
    • Meetings: a microphone set to another sample rate than the speakers (such as 44.1 kHz) recorded nothing and kept restarting; it now records at its own rate, and when a microphone still records nothing, Dicta turns on Echo cancellation for that recording.
  3. 0.1.0-beta.7Beta

    Fixed

    • Fixed a crash when logs contained malformed links: a link Dicta could not read (such as one with a broken % escape) in what it writes to its log could make it quit. Such links are now logged safely, with any password or key in them still hidden.
  4. 0.1.0-beta.6Beta

    Added

    • Tasks at once (Settings → Transcription; 1 by default, up to 3): how many links download, and how many files are prepared, at the same time; the others wait in line (“Queued · #2”). Transcription itself still runs one at a time.
    • Clean view: the eraser button above a transcript (or Settings → Transcription → “Clean view”; off by default) hides filler words (um, uh, 嗯, 呃) and words said twice by mistake, by rules on this Mac, in Chinese, English, Japanese, Korean, Spanish, French, German, Russian, Portuguese and Italian. The transcript itself does not change; copying and exporting from the page follow what it shows. Numbers, units and abbreviations are never removed.

    Changed

    • Less disk space: a file's or a link's audio is now kept as Apple Lossless instead of WAV (the same sound, about 30–70 MB instead of 115 MB per hour of speech), and a phone recording as the phone sent it. Settings → Advanced → “Compress old recordings” compresses older ones, losslessly. With “Keep the video” on, a link's video is kept at 720p or less when the site offers it.
    • Faster search in large libraries, a smaller library file: searching one or two Chinese characters is about three times faster in a 500-hour library. The first start after this update converts the library once, and Dicta takes a little longer to open (about 15 s for 500 hours of recordings).
    • Less memory while idle: models leave memory after 5 idle minutes, and Dicta's engine processes quit when they have nothing to do and start again with the next task (Settings → Models & network → “Unload idle models”; 0 keeps them). The live-captions model unloads as soon as a meeting ends.
    • Times as on the transcript: agents, chat citations, command-palette results and live captions write a moment as the transcript does (“04:05”).

    Fixed

    • Meetings: a paused meeting no longer starts recording again by itself after the recording engine crashes; stop timers and reminders that passed while the Mac slept now come within a minute of waking; Stop ends a meeting within about half a minute even when the engine does not answer, keeping what was recorded; a meeting started just as Dicta opened is no longer lost; a meeting saved after a crash gets its automatic export and email draft too.
    • Live captions run in a process of their own: a crash no longer interrupts the recording, and captions come back within a few seconds. They skip silence (about half the cost over a mostly quiet meeting), try again when their model could not load, and no longer replace earlier captions after engine restarts.
    • Transcription and AI: automatic titles and summaries no longer stop for good after the engine kept crashing on one file; a transcription stuck at “Queued” can be stopped and resumes at the next start; a long summary that ran out of room is written again or in parts, or marked “May be incomplete”; summaries keep years such as “2025年”; asking about a long meeting keeps its end; a model too large for this Mac's memory says so instead of crashing the engine.
    • Dicta froze on some Markdown: an AI answer, a summary or a note with a list followed straight away by a heading, a line, a quote or a code block. It now shows them.
    • Library: searching a large library no longer slows typing, and editing a long transcript is quicker; a speaker with nothing left to say is removed from the library, People and search by person; “Archive after N days” counts from when a transcript was transcribed or brought back; space left by deleted recordings, crashes and older versions is freed a minute after Dicta starts; moving the data folder no longer rewrites its name inside transcripts and notes.
    • Links and watched folders: when YouTube is only limiting downloads, Dicta says so and tries again later instead of “The video is not there any more”; a full disk is reported as such. In a watched folder, a new file named like one already transcribed and moved away is transcribed, a renamed file is not transcribed again, nothing is imported twice after a quit, and unchanged folders are checked less often.
    • Updates download through the proxy in Settings → Models & network (by default the system's), and a broken download goes on from where it stopped.
    • Phone: a recording that arrived just before this Mac stopped is now transcribed when the phone checks in or Dicta starts again; phones whose names have quotes can pair, and “Connected” shows only once the phone is kept; Connect a Phone shows the QR code first, with the six-digit code under “Can't scan?” (allow only if both screens show the same check code); other computers on the same network can no longer lock the pairing code or keep a paired phone waiting, and web pages cannot use it.
    • Exports: an edit made just before quitting reaches the exported files, date-first names no longer change with the time zone, and a meeting's audio exports about 20% faster. Agents: search_transcripts pages no longer repeat or skip a transcript, and Dicta no longer quits when Settings → Agents changes as an agent connects. Voice profiles: “not this person” that cannot remove a voice now says so and changes nothing. Thai, Lao, Khmer and Burmese subtitles keep their spaces and keep vowel and tone marks with their letters; dictionary words apply in Thai and Lao.
    • Built with Go 1.27.4 (security fixes in certificate checks and HTTP).
  5. 0.1.0-beta.5Beta

    Added

    • Hide the Dock icon: Settings → General → “Show Dicta in the Dock” (on by default). Turned off, Dicta is not in the Dock or ⌘Tab; open it from the menu bar icon, which then stays on. Shortcuts work as before, but macOS shows no menu bar for the app.
    • Settings → Phone (off by default): pair this Mac with Dicta Recorder on your iPhone (the iPhone app is in beta testing, not yet on the App Store). On the same Wi‑Fi, the phone's recordings come to this Mac, are transcribed and summarized, and the results go back to the phone. Only while it is on does Dicta listen on the local network, and it accepts only paired phones' encrypted requests; macOS asks for Local Network access.

    Fixed

    • Meetings that would not start on some Macs: when another app or a headset changed the microphone's format as a meeting began, the recording engine crashed again and again and the meeting ended with “No sound was heard”. The microphone now starts again with its new format; if the engine still keeps crashing, the meeting stops after the third time and says why.
    • “Not enough free disk space” with room to spare: Dicta now counts free space as Finder does (with what macOS frees on demand), so it no longer refuses to record or to download a model while Finder shows plenty of room. When the disk really is low, the message says how much is free and where to clear some.
  6. 0.1.0-beta.4Beta

    Added

    • See the text while a file is transcribed: the transcript page fills in as transcription goes (Standard transcription a sentence at a time, Fast every few seconds), with times to click and, for Standard, speakers. It is a preview: punctuation, speakers and paragraphs are tidied at the end, and the page then switches to the finished transcript where you were.
    • Stop timer for meetings and recordings: click the clock in the meeting window (or Stop timer in the recording dialog and on the Meetings page) to stop in 15–90 minutes, at a time, or when the calendar event ends. A minute before, you are told and can extend by 10 minutes, cancel or stop now; it never stops without that warning. Settings → Meetings → Stop timer can also end calendar meetings a few minutes after their event (off by default) and reminds you after a long recording (3 hours by default; it only reminds). Links take stop_after=30m or stop_at=15:30.
    • Voice profiles (Settings → Voice profiles, off by default): name a speaker once (Set person in the speaker's ⋯ menu; for anyone but you, confirm they agreed) and Dicta suggests that person in your other recordings, “Maybe Zhang San · 82%”, with ✓ and ✗. Nobody is named without you unless you turn on Name speakers when sure. Voices are made and kept only on this Mac: never uploaded or synced, and never in exports, Refto or what agents read (names only). Turning it off offers to delete all voice data.
    • People: with voice profiles, a People page (under Tools) lists the people you named, with their recordings, talk time and share of each; All and the archive filter by person, and agents can search by person (names only). Only speakers you confirmed count.

    Changed

    • One person stays one speaker in long recordings: in recordings over 10 minutes, someone who was quiet for a while no longer comes back as a new speaker. Dicta compares the voices on your Mac while transcribing and keeps none of them (on by default; Settings → Transcription → Advanced). It needs the Speaker matching component (14 MB), part of the recommended models.
    • Meetings take less than half the disk space: one compressed recording, and each side in a lossless format about 40% smaller (the same sound, sample for sample). A side that heard nothing (nothing played, or only a quiet room) is no longer saved on its own or transcribed, so the meeting is ready sooner. Meetings recorded before keep their files.
    • When a meeting goes quiet: after 10 minutes of silence the meeting window asks “Is the meeting over?” (End recording or Keep recording; it never ends by itself; Settings → Meetings sets the minutes or turns it off). After 3 minutes without speech, live captions unload their model (about 700 MB) and load it again when someone speaks, without losing words.
    • Shorter labels and prompts: buttons, menus and status lines no longer end with “…”; the meeting window's prompts are one small line with icon buttons.
  7. 0.1.0-beta.3Beta

    Added

    • Vocabulary sets: Templates has a third kind, Vocabulary: the names and terms of one topic (a podcast, a course, a client), one a line. Choose one when you transcribe on the New page, for a batch or a watched folder (each folder keeps its own), or when you transcribe again; meetings and recordings have their own in Settings → Meetings (none at first). A set adds to the general vocabulary (New page → Advanced options), which every transcription uses; live captions use them too. The transcript's Info shows the set it was made with. Standard and Fast transcription use the vocabulary; multilingual transcription (Whisper) does not.
    • Agents (Settings → Agents, off by default): let your own AI agents — ChatGPT's Codex, Claude, Cursor, WorkBuddy and others — search and read your transcripts, their summaries and notes over MCP, on this Mac only, and, with a second switch, start transcriptions of files and links. Dicta asks you before each program's first use and first transcription, lists the programs allowed (remove one at any time) and their recent calls (never what they asked or read); one click adds Dicta to a client and installs Dicta's skill for agents. What an agent reads goes to its AI provider; Dicta itself still sends nothing.
  8. 0.1.0-beta.2Beta

    Fixed

    • Save to Refto (formerly Save to Act) works again: Act is now called Refto, and in 0.1.0-beta.1 saving there did not work. With Refto installed, Save to Refto in the export dialog and Save the Recording to Refto in the More menu add the export or the recording to your Refto library; an old copy of Act is no longer offered.
  9. 0.1.0-beta.1Beta

    The first version: meetings, recordings, files and links transcribed and summarized on your Mac, with local models only.

    Added

    • Meetings, without a bot: one button records every app's sound and your microphone (or one app, or the microphone only), with a small window over every app that has the clock, captions as people speak, Pause and Ask AI. When a call app starts using the microphone, Dicta starts recording 2 seconds later and tells you (calls in a browser ask first; it can also always ask, or not watch at all). When the meeting ends it is transcribed with speakers and summarized; exporting its files and drafting an email are a switch away. Connect your calendar, and meetings are named after their event and the summary knows who was invited.
    • Recordings: record your microphone from New, the Meetings page or the menu bar; pause, or hide the dialog and keep recording; it is transcribed when you stop.
    • Live captions: captions of what the Mac plays (a video, a talk, a call you listen to) or of the microphone, translated into your language by the local AI if you like, in a caption bar over every app that can be hidden from screen sharing; ask "What did I miss?" while it runs, and keep it as a transcript when you stop.
    • Batch & watch: transcribe many files at once, or every new file in a folder you watch, each exported when it is done.
    • Files and links: drop audio and video files (or choose them), or paste a link (YouTube and other sites: a link the site refuses is tried again in other ways, and the downloader keeps itself up to date); long recordings of several hours are transcribed in windows, with speakers kept apart across them.
    • Transcription on the Mac: Standard transcription with speakers (MOSS-Transcribe-Diarize), Fast transcription (Qwen3-ASR) and multilingual transcription, picked by language, all with MLX on Apple silicon; nothing leaves the Mac.
    • The transcript: a player with the waveform and speaker lanes, the video beside the text, speakers to rename and merge, edits, bookmarks, tags, search across the library, and export to SRT, VTT, Markdown, CSV, JSON, text and PDF (exported files follow your later edits); with Act installed, an export or the recording can be saved to your Act library.
    • AI on the Mac: summaries written as a transcription finishes (Gemma 4), templates of your own, chat about a transcript, Ask across the library, and translation of the transcript.
    • Models: the best models by default, downloaded once, with tabs for transcription, AI and the rest; a proxy and a mirror for downloads. They are kept in ~/Library/Application Support/ActNow/models, shared with the other ActNow apps; models already in the Hugging Face cache are brought over without downloading them again, and Time Machine leaves models out unless you turn backups on in Settings → Advanced.
    • Links ask first: a dicta:// link from a web page or another app that starts, stops or discards a recording asks before it does.
    • Updates inside Dicta: a daily check; a new version shows at the foot of the sidebar, and its panel has the notes in your language, downloads only what changed with its progress and restarts to finish when you say so. Check for Updates… in the Dicta menu, the command palette and Settings → About, which lists every version's notes and has "check automatically". Early-access users are told before an update that needs a paid licence, and can keep the version they have.
    • English and Simplified Chinese; signed with a Developer ID and notarized.
    • One window in the ActNow family's layout (as Act's): a search field over the sidebar (⌘F finds recordings by title and what was said; ⌘K inside it), the pages most people use at its top, tags and tools in sections that fold, the archive at its end; what is going on (a meeting's clock, transcriptions, downloads, a stopped engine) as cards at its foot, over Settings, the logs and the notifications; "View all" opens Tasks. Settings come in fewer sections, as groups with a line under each setting; the models are in Settings → Models & network.
    • Sounds (Settings → General → Sounds, off by default): a sound when a task finishes (a transcript, a download, an export) and another when one fails, each chosen from the ActNow family's sound library (Bloom and Shrug by default), macOS's system sounds or none, with a compact picker that searches them (names and tags), filters by type and plays each with ▶ (again to stop); a volume in steps.
    • Activation: Dicta asks for the licence code from your email or your ActNow account the first time it opens (early access: request it on actnow.dev). The licence works offline and checks in with actnow.dev once a week; Settings → Licence shows its edition, when it ends and the Macs it is on, opens actnow.dev to manage them, and changes the code. Until Dicta is activated, meetings are not detected or recorded and links wait; when a licence ends, your recordings, transcripts and settings stay as they are.