Download icRec

spctl -a -vv -t open --context context:primary-signature ~/Downloads/icRec-0.1.67.dmg
shasum -a 256 ~/Downloads/icRec-0.1.67.dmg

• If the first line prints “accepted” and “source=Notarized Developer ID”, the file is notarized by Apple.
• If the value from the second line matches the SHA-256 in the table above, the file has not been altered.

Notarization is Apple's automated check for known malware. It does not guarantee the app's quality or behavior.

When a new version is out, “Update available” appears at the bottom of icRec's sidebar. Click it to read the release notes, then press “Update and restart”: icRec downloads the new version, checks the file's checksum, Apple notarization and the publisher's signature, then replaces itself and restarts. Recordings, text and settings are kept.

• icRec checks at launch and once a day. You can also check now under Settings → Updates, and turn automatic checks off there.
• If icRec runs from somewhere other than the Applications folder, it can't replace itself; download it from this page and replace it by hand.
• Version 0.1.0 has no update feature; if you use 0.1.0, replace it from this page.

See Updates in the user guide for details.

• Fixed “Codex を自動検出” (Detect Codex) in Settings saying Codex CLI could not be found even though the ChatGPT app or Codex was installed. icRec now also finds Codex installed with its standalone installer (~/.local/bin) and the Codex bundled with the current ChatGPT app.
• Fixed “アップデートして再起動” (Update and restart) getting stuck at “アップデート中” (Updating), with “Quit icRec” greyed out and the new version never installed. If this happens in 0.1.65 or earlier, force quit icRec with ⌥⌘Esc (Option + Command + Esc) and install the new version from this page. From this version on, updates install automatically.
• When Codex was set to a model your ChatGPT account can't use, summaries, “文章を整える” (Tidy up the text) and other AI work failed with a “model is not supported” error. icRec now switches to Codex's default model in that case. The model shown under “AI 機能” (AI features) in Settings is the one actually used.
• Added “このアプリについて” (About this app). Open it from the icRec menu (or the menu bar icon’s right-click menu) to see the version, the copyright (© 2026 K386) and a link to the official site. It appears once on first launch and once after each update.
• Added “タスクを洗い出す” (Find tasks) to “この日のまとめ” (Summary of the day) in the Calendar. AI finds to-dos in the day’s report and lists them under it with checkboxes; add the ones you choose to Tasks (the same report in Notes shows them too).
• When LM Studio is your AI, it now works without opening LM Studio first. icRec starts LM Studio's server when icRec starts and, just before a summary or other work, loads the chosen model if it isn't loaded. A model icRec loads is unloaded after an hour unused. Turn this off with “LM Studio を自動で起動する” (Start LM Studio automatically) in Settings.
• Voice memos can now also record a meeting app such as Zoom (macOS 14.2 or later). Turn on “会議などを録音する” (Record meetings and the like) and choose the app: icRec takes its sound directly, rather than picking up the speakers with the microphone, and records it with your voice in one recording. When you stop, both are leveled automatically before saving. The first time, macOS asks for “System Audio Recording” permission.
• Long texts in a recording’s details, such as MindClip transcripts, now fold to eight lines. “Show full text” expands them and “Collapse” folds them again. Texts containing the search term open in full.
• Fixed “Ask icRec” infographics and similar failing with “Expected integer, received float” on LM Studio (the output cap was computed as a fraction). Slides, mind maps and infographics were verified with LM Studio (Qwen3.6).
• Fixed “Find tasks” holding back candidates as “unverifiable evidence” when the AI quoted a passage that runs over a line break as one line. Whitespace is now ignored when checking that a quotation is in the original.
• Fixed waits of up to 10 minutes when a model in LM Studio kept repeating itself instead of finishing its JSON. Output is now capped (sized to the input) and the request is retried once when the cap is hit. Requests to OpenAI, Claude and Gemini also carry an output cap (16,000 tokens).
• Added LM Studio (a local LLM) to the API key option under AI. With a model loaded in LM Studio on your Mac and its server started, everything Codex does (summaries, tidy-up, tasks, Ask icRec, daily reports) works for free with no text leaving the Mac. Choose “LM Studio（ローカル）” in Settings and pick the model with “接続を確認” (Check connection). The model must support structured (JSON) output.
• Added “API キー” (API key) under AI in Settings. With an API key from OpenAI (ChatGPT, Codex), Anthropic (Claude) or Google (Gemini), you get everything Codex does — summaries, tidy-up, task extraction, Ask icRec and daily reports — without installing Codex. Choose the company and model in Settings; the key is kept in the Mac’s keychain. Usage is billed at each company’s API rates.
• Fixed transcription stopping with “長い録音を分割できません” (the long recording could not be split) for Mindclip recordings longer than two hours. Press “文字起こし” (Transcribe) again on such a recording to process it.
• When recording with your Mac, icRec now warns during recording if the chosen microphone sends no sound at all. If you stop anyway, it tells you which microphone was silent and suggests another one, instead of “the transcript was empty” (a recording with no sound is not transcribed).
• “文章を整える” (Tidy up) with Apple Intelligence no longer silently drops what was said when the AI summarizes a passage instead of tidying it. Such a passage is tried once more, and if it is summarized again it is kept as it was. The activity list shows how many passages were kept.
• Fixed “文章を整える” (Tidy up) with Apple Intelligence sometimes failing on long recordings with “Failed to parse generated content.” Retry the failed item from the activity list to tidy it up.
• Connecting to Whisper Local now takes one confirmation. Click “Whisper Local と連携する” (Connect to Whisper Local) under Transcription in Settings, then “連携する” (Connect) in Whisper Local: the client ID, token and folder ID are set up for you (needs Whisper Local 0.3.15 or later). Without Whisper Local, Settings says it is needed for transcription and where to get it.
• Settings now has an “AI” section where you choose which AI to use: none, Apple Intelligence, or Codex (ChatGPT). On an Apple silicon Mac with macOS 26 or later, Apple Intelligence summarizes, tidies up and lists tasks for free, without sending text off the Mac. The chosen AI is shown in the sidebar (green when ready). Home’s “Ask icRec” (answers, slides and more) and daily reports need Codex; without it, Home is a keyword search, and “What is Codex?” explains what you need (a paid ChatGPT plan and the ChatGPT app).
• The table of contents at the top of Settings is easier to use: it now has a “設定の目次” (Contents) title, lines up with the sections below, and highlights the section you are looking at.
• Settings now shows links to each section (storage, import sources, transcription, tags, Google and more) at the top. Click one to scroll that section to the top of the window; the links stay visible while you scroll.
• Added “Launch Whisper Local in the background when icRec starts, if it isn’t running” under “Transcribe automatically with Whisper Local” in Settings (off by default). You no longer need to set Whisper Local to open at login; if it can’t be found or started, the connection status in Settings says so.
• In Home’s “Ask icRec”, “Instruction for AI” under the search box now offers slides, a mind map and an infographic besides an answer. icRec finds the recordings, notes, daily reports and tasks related to the question and builds it from them. Slides (a cover and 4–8 pages) export as PDF; mind maps and infographics as PNG images. Any of them can be saved to Notes or copied as text with their sources; the [numbers] in the text point to the sources listed below. After a result, choosing another instruction remakes it from the same question. Requires Codex; the question and excerpts of the related items are sent to OpenAI (never all recordings).
• Transcripts can now be tidied up automatically (“Tidy up after transcription” under “Transcribe automatically with Whisper Local” in Settings). Whenever a transcription finishes — a Mac recording, an import from a recorder or a connected service, a manual or batch transcription, or a retry — icRec runs “Tidy up” on it. When several recordings come in together, all are transcribed first and then tidied up one by one. The original transcript is kept. Transcripts already tidied up and very short ones are skipped. If tidying up fails, the transcript stays and you can retry from Activity. Requires Codex; when on, each transcript is sent to OpenAI and uses your ChatGPT allowance (audio is never sent).
• Fixed transcripts filling up with lines like “ご視聴ありがとうございました” (“Thanks for watching”). Whisper writes these on silent or noisy stretches of a recording, so icRec now removes them from transcripts. It happened most with recordings left running, such as MindClip. Existing transcripts are unchanged; transcribe a recording again to clean it.
• On the HiNotes sign-in page, choose “Remember my email and password on this Mac” and they are filled in next time. HiNotes allows only one sign-in at a time, so signing in with the HiNotes app signs icRec out; signing in again now takes only the image check and “Log in”. They are kept only in the Mac’s Keychain and can be removed anytime with “Delete remembered sign-in” in Settings.
• Fixed the HiNotes sign-in window staying open with “HTTP 405” after signing in. Please sign in again with “Sign in to HiNotes” in this version.
• HiDock recordings can now be imported automatically from HiNotes (Settings → “Import HiNotes recordings automatically”). Sign in with your email and password on the HiNotes page opened by “Sign in to HiNotes”, and while icRec is running it checks HiNotes every 5 minutes for new recordings and whispers and imports them with the “HiNotes” tag. The recording date is the one HiNotes recorded, and the name is the one HiNotes gave. Only the audio is imported; icRec transcribes it when transcription is on. Your password is not stored in icRec, and recordings in HiNotes are never changed or deleted.
• Recordings longer than two hours can now be transcribed. Whisper Local accepts up to two hours at a time, so a long recording is split automatically at pauses in speech, each part is transcribed, and the results are joined into one transcript. The recording stays a single item, and caption formats (SRT, VTT and so on) keep times counted from the start of the recording. A recording that failed with “files up to 2 hours are supported” can be transcribed with Retry in Activity.
• Fixed MindClip recordings never being imported. With this version, recordings made after turning the option on are imported; earlier recordings that were not visible when it was turned on may be imported too. Delete any you don't need in icRec (deleted recordings are not imported again). The tutorial now lists MindClip among the ways recordings come in.