# TypeWhisper for Mac: on-device dictation, meeting automation and a paid tier > TypeWhisper for Mac transcribes speech with on-device models or through Groq, OpenAI and xAI, then applies workflows, dictionary terms and meeting sidecars. The Swift app pairs an Xcode project with a CLI directory and no documented build command, while its Premium tier sits beside a copyleft grant. **TypeWhisper/typewhisper-mac** — Local speech-to-text for macOS on-device AI, fully private, optional cloud - Repository: https://github.com/TypeWhisper/typewhisper-mac - Stars: 1,874 · Forks: 148 - Language: Swift - License: GPL-3.0 - Published: 2026-10-14 · Updated: 2026-10-14 · Language: en - Canonical page: https://hysenlabs.com/projects/typewhisper-typewhisper-mac ## Dictation into any text field without shipping your voice away TypeWhisper for Mac is a Swift application for dictation, file transcription and meeting notes, built for people who would rather speak a paragraph into an editor than type it. The badges pin the floor at macOS-14.0%2B and Swift-6, and the topic tags `apple-silicon`, `dictation`, `on-device`, `privacy`, `swiftui` and `whisperkit` describe the shape of the thing. The surface area is far wider than a dictation hotkey. Screenshots in the repository cover a Home dashboard, Recorder, History, Dictionary, Snippets, Hotkeys, Workflows, File Transcription, an Integration Marketplace, Statistics, and General plus Advanced settings. Separate directories add a widget target (`TypeWhisperWidgetExtension/` and `TypeWhisperWidgetShared/`), a plugin stack (`PluginRegistry/`, `Plugins/`, `TypeWhisperPluginSDK/`), a transcribe action (`TypeWhisperTranscribeAction/`) and a command line binary (`typewhisper-cli/`). The `.github/` tree also carries an ISSUE_TEMPLATE for per-candidate ship checks, which is how this project tracks what a build is allowed to claim. ## Two transcription paths decide where your audio lands Transcription runs either on-device or through a cloud API, and that choice belongs to your configuration rather than to whoever compiled the build. Groq, OpenAI and xAI/Grok are the named remote providers, and the 1.6 feature notes add ordered LLM provider fallbacks, explicit API-driven dictation and a clearer active-provider status, so you can see which engine produced a given transcript. Audio enters through a microphone or through files on disk. Recording startup moved off the main actor, eligible microphones are prewarmed, and microphone priorities plus route recovery were reworked, with controlled tests measuring lower request-to-first-buffer times for built-in, USB and AirPods inputs. Those tests publish no figures anywhere in the visible README, so the latency claim stays uncheckable from the repository itself. Nobody reading the docs can tell you the first-buffer time on a given Mac. ## Fetching the app: badges and tags but no build command There is no installation command in the visible README, no package manager name and no download link, so there is nothing to paste into a terminal. What it offers instead is a requirements pair in the badges and a repository layout that fits a source build: `TypeWhisper.xcodeproj/`, `CodeSigning.xcconfig`, a `fastlane/` directory and a `scripts/` directory. The tags explain themselves too. `v1.6.1` was cut on 2026-09-16, and the two newer tags, `v1.7.0-daily.20260918` and `v1.7.0-daily.20260917`, carry a dated suffix instead of a version bump, the shape of an automated daily build rather than a hand-cut milestone. The last push was on 2026-09-20. A reader who wants a stable line should pin 1.6.x and ignore the daily tags, because the README itself still calls 1.6 the current stable build for macOS while the 1.7.0-daily tags sit at the top of the list. ## Live transcript text versus the authoritative final insertion Dictation into another application runs in two stages, and the difference matters when a long sentence lands badly. Live transcript updates can write into the active text field while you are still speaking; final insertion then replaces that provisional text with the authoritative version. A half-finished pass therefore appears on screen and corrects itself, which is a deliberate answer to the question of who wins the race between streaming text and the finished one. Three indicator styles sit above that flow: Notch, Overlay and Minimal, each of which gained better placement, visibility, interaction and screen-capture controls. Route recovery sits beside them, which tells you the project treats a dead input or a hotkey firing into the wrong window as a real failure mode rather than an edge case. That is the failure that actually ends a dictation habit, and it is the one the screenshots bother to document, with a dedicated Recovery screen. ## Backup and Restore enumerates data kinds but no file format Portable Backup and Restore arrived with the Settings redesign, and its scope is spelled out in one long clause: workflows, dictionary entries, snippets, profiles, prompt actions, hotkeys, installed community plugins, history text and supported preferences. What the repository does not spell out is the format. No archive extension, no manifest, no destination path, no partial-restore behaviour. That gap matters more than the list does. A backup you cannot inspect is a backup you cannot verify before moving to a second Mac, and nothing in the visible docs lets you script that check. The same silence covers the dictionary and snippet side of the app, where a `termpacks/` directory exists with no published schema. A support matrix page is linked for the shipped feature set, yet nothing states which of those nine data categories survives a restore that stops halfway. ## Local meeting checks open Recorder sessions with Markdown sidecars Meeting automation is the part that reaches past dictation. The 1.6 notes describe optional reminders, countdown controls, local meeting-activity checks, automatic Recorder sessions, sortable filenames, transcript metadata and Obsidian-ready Markdown sidecars, aimed at Zoom, Microsoft Teams, Google Meet and FaceTime. Obsidian live sync and templates extend the same path for people who keep notes in a vault. Local meeting-activity checks deserve careful reading. The trigger is a check on the Mac rather than an upload of a calendar account, but the feature still depends on recognising that a conferencing app is running and then routing its audio into the Recorder. Sortable filenames and a sidecar give you a tidy archive shape, yet no retention policy is documented for the history text those sessions accumulate. A diarist who records every call inherits a cleanup problem the app does not describe. ## Four extension surfaces, and the CLI that documents none of them The tree exposes four extension points and describes one of them in prose. The workflow side exposes `TypeWhisperPluginSDK/`, `PluginRegistry/` and a `Plugins/` directory, with bulk plugin updates and an Integration Marketplace screen for installing community plugins. A macOS action target, `TypeWhisperTranscribeAction/`, lets a shortcut pipeline call transcription. A `typewhisper-cli/` directory points at a command line entry point, and no flag, subcommand, exit code or config key for it appears in the visible documentation. That last gap is the closest thing to a real alternative in this repository, because it is the only architectural contrast the tree supports: a GUI application with indicators, insertion rules and a settings surface on one side, an undeclared binary on the other. The project names no rival application, so any comparison against macOS own dictation would have to come from outside these docs. For tooling, the `AGENTS.md`, `CONTRIBUTING.md`, `SECURITY.md` and `TRADEMARK.md` files plus `.markdownlint-cli2.yaml` explain the Ruby pipeline, `Gemfile.lock` and `fastlane/`, documented in `docs/screenshot-automation.md`. ## Premium Cloud Sync sits beside the fully private promise The repository is GPL-3.0, and the tree carries `LICENSE` and `LICENSE-COMMERCIAL.md` as separate files, so the copyleft grant covers one half of what the project ships. The screenshot set makes the other half plain: Premium Overview, Premium Access, Premium Calendar, Premium Correction Learning, Premium Cloud Sync and Premium Locked name a paid tier with correction learning and sync, and no README text says which features the lock covers. That is the contradiction worth holding onto. The project description calls the app fully private with optional cloud, the topics carry `on-device` and `privacy`, the Settings redesign ships privacy-friendly usage insights, and a Premium Cloud Sync panel sits inside the same binary, alongside an MCP Client action that talks to external tools. Privacy here is a setting you choose, not a property of the build. Nobody outside the project can check whether correction learning and cloud sync move dictionary entries, snippets or transcripts off the machine, and `TypeWhisperICloudBridge/` in the tree is the only trace of a sync path. ## Conclusion TypeWhisper suits a Mac user who wants dictation, file transcription and meeting notes without uploading audio by default, and who is willing to build from an Xcode project because no install command is documented. It is the wrong tool for anyone who needs a scriptable command line today, since `typewhisper-cli/` publishes no flags, or for anyone who expects every feature to be free, since the screenshot set names a locked Premium tier. Before adopting, settle the dual licensing question for commercial use, confirm your machine clears macOS 14.0+, and decide whether a cloud sync panel belongs in an app sold as fully private. ## FAQ ### Does TypeWhisper for Mac upload my voice recordings to a server? Not by default: audio can be transcribed by on-device models, and the README also names Groq, OpenAI and xAI/Grok as cloud options. The path is a configuration choice, and the 1.6 notes add a clearer active-provider status so you can see which engine is in use. ### What are the macOS requirements for TypeWhisper? The project badges state macOS 14.0+ and Swift 6, and the topics tag Apple Silicon. No CPU, memory or model size requirement is given in the visible documentation. ### Is TypeWhisper free open source software? The repository is GPL-3.0 and the tree also holds LICENSE-COMMERCIAL.md, so commercial use is a separate question from the copyleft grant. Screenshots include a Premium Locked screen, and the README does not list which features sit behind it. ### Can TypeWhisper transcribe an existing audio file? A File Transcription screen and a Recorder screen both appear in the screenshot set, and meeting automation can open automatic Recorder sessions with sortable filenames, transcript metadata and Obsidian-ready Markdown sidecars. No supported input format is documented. ### How do I install TypeWhisper and set a dictation hotkey? The visible README documents no install command, so a source build through the Xcode project is the only path it points at. A Hotkeys screen exists in the screenshot set and hotkeys are carried by Backup and Restore, but no default key combination is printed anywhere. ## Sources - [Issues](https://github.com/TypeWhisper/typewhisper-mac/issues) - [License: GPL-3.0](https://github.com/TypeWhisper/typewhisper-mac/blob/main/LICENSE) - [README](https://github.com/TypeWhisper/typewhisper-mac/blob/main/README.md) - [Releases](https://github.com/TypeWhisper/typewhisper-mac/releases) - [TypeWhisper/typewhisper-mac on GitHub](https://github.com/TypeWhisper/typewhisper-mac) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/typewhisper-typewhisper-mac