Mac / iPad / iPhone App

Kakitome

An AI transcript editor. Record the meeting, transcribe it, finish the document.

Kakitome is a text editor with transcription and AI built in. Press record and the words appear as you speak. Fix the names speech recognition keeps getting wrong with a replacement dictionary, run a preset to summarize or proofread, and drop the result straight into your document. From audio to finished text, in one window.

🔒 Transcription and summarizing both run on your own device. Neither the audio nor the text is sent anywhere — including to the developer.

🌐 The interface comes in English and Japanese. It follows your device, and you can also set it per app under Settings › General › Language.

View privacy policy

Record, and watch it become text

Words appear as you speak

Start recording and transcribed text is appended every few seconds, so you never have to wait until the end. Committed sentences can be corrected on the spot, mid-recording, and your corrections are what get saved.

Audio and video files too

Pick a wav, mp3, m4a, aac, aiff, flac, mp4 or mov file and transcribe it. Progress and an estimate of the time remaining appear in the status bar.

Choose the shape of the output

Plain prose, one sentence per line, or timestamped ([00:01:23]). Turn on speaker separation and a heading is inserted for each speaker.

No phantom filler words

Speech recognition tends to invent short interjections over silence. Kakitome checks the confidence of each segment and drops them automatically — and you can turn that off.

Four transcription engines

Apple speech recognition (default)

Built into macOS and iOS. Nothing to download, fast, and good with long audio — so it works the moment you install the app.

NVIDIA Parakeet (Japanese)

Strong on messy recordings such as meetings and interviews, at over 10x real time. About 700 MB.

Cohere Transcribe

The most accurate Japanese of the four, at about half of real-time speed. For when accuracy matters more than the wait. About 1.9 GB.

Whisper

From Large v3 Turbo down to Tiny. Steady on clean, read-aloud audio (Mac only).

All of them run on your device. Only the model weights are fetched, from Hugging Face, when you choose to download one in Settings.

Turn it into something readable

Replacement dictionary

Register the company names, product names and people that speech recognition keeps getting wrong, and they are all replaced in a single pass. The longest match wins, so the order of your entries never matters. Regular expressions and whole-word matching are available per entry.

AI presets

Translation, proofreading, summarizing, meeting minutes, polite and plain Japanese style, plain-language rewriting, Markdown cleanup, image and video generation prompts. Results appear in a lower pane, so you check them before they touch your document.

Meeting minutes

Sorts a transcript into decisions, action items and open questions. If the recording is matched to a calendar event, its title and attendees are passed along as context.

Three AI engines

Apple Intelligence (nothing to set up, offline), built-in MLX models that run on your device's GPU, or a local LLM / OpenAI-compatible API you host yourself — Ollama, LM Studio, llama.cpp server.

Rule-based conversions

Full-width to half-width, whitespace and blank-line cleanup, joining wrapped lines, sorting and deduplicating. Strip Markdown, Markdown to HTML, lists, quotes, headings, TSV to Markdown table, JSON. Offline, instant, undoable with ⌘Z.

Shortcut bar

Put the functions you actually use along the top and run them with one click. Right-click to add, drag to rearrange, three button sizes.

Tied to the meeting it came from

A monthly calendar

Your device's calendar events sit in the top right. Days with events get a dot, days with a transcript get a waveform. Double-click a date to open everything recorded that day.

Recordings matched to events

Kakitome looks for an event that overlaps the time you were recording and puts its title, time, location and attendees at the top of the transcript. The event title is added to the recording's filename too.

Written back into the event

A one- or two-sentence summary is appended to the event's notes, leaving whatever was already there intact. Months later, the calendar alone tells you what was discussed.

Google Calendar too

Included as long as you have added the Google account in your system settings. Kakitome never talks to Google directly — it only reads the calendar your OS already has.

And it is still an editor

Kakitome opens .txt and .md files. Alongside UTF-8 it detects Shift_JIS, EUC-JP and ISO-2022-JP, and always saves as UTF-8. Line numbers, find and replace, text size and a monospaced option are all there. Mac and iPad use two columns — editor on the left, calendar and buttons on the right — and iPhone a single column.

Price

Free to use. That includes the editor, the replacement dictionary, the formatting and output conversions, running AI presets by hand, transcription with no engine restrictions, and the monthly calendar.
The only limit on the free version is 15 minutes per recording or transcription.
There is a single one-time purchase, Kakitome Pro. Not a subscription. It removes the length limit and enables automatic summaries, calendar integration, copying the transcript .txt to a folder of your choice, and speaker separation.
One purchase covers Mac, iPad and iPhone on the same Apple ID.
There are no ads.

Requirements

macOS 26.2 or later on an Apple silicon Mac, or iOS / iPadOS 26.0 or later on iPhone and iPad. The Apple Intelligence engine needs Apple Intelligence turned on in system settings; the rule-based conversions work anywhere. Built-in models use about 3 GB of memory for a 4B-class model on a Mac, and about 1.1 GB for Bonsai 4B on iPhone and iPad.

View privacy policy