MacParakeet Review 2026: Free Local Dictation for Mac

Direct answer — Is MacParakeet worth using in 2026? MacParakeet is a free, GPL-3.0 macOS app that runs speech recognition locally on Apple Silicon. Version 0.8.4 covers system-wide dictation, file and URL transcription, and meeting recording with speaker labels. It requires macOS 14.2 or later and an M-series chip, and there is no Windows build. For a Mac-only workflow, it is a genuinely capable free option.

Most MacParakeet review searches come from the same place: someone has found a free, open-source Mac dictation app and wants to know what the catch is. That is a fair question, because free local dictation tools usually pay for their price tag somewhere — in language coverage, in stability, or in the amount of tinkering they demand.

This review covers what MacParakeet actually does, what the GPL-3.0 licence gives you, where its language support genuinely stops, and how it compares with a paid cross-platform tool. We publish Weesper Neon Flow, so treat the comparison section accordingly — the goal here is to be specific about where each tool wins rather than to declare a winner.

What is MacParakeet, and what does it do?

MacParakeet is a local-first voice app for Apple Silicon Macs that bundles dictation, file transcription and meeting recording into one application. Its own repository describes it as a “fast, private, local-first voice app for Apple Silicon Macs”.

The feature set is unusually broad for a free tool:

• System-wide dictation — press a hotkey, speak, and text appears wherever the cursor is
• File and URL transcription — drop in an audio or video file, or paste a link to a YouTube video, a Loom recording or a podcast episode
• Meeting recording — capture system audio and the microphone together, or either one alone
• Speaker labels — transcripts include word-level timestamps and speaker labels when the selected engine provides timings
• Live meeting notes — type during the call with slash commands, auto-saved
• Custom vocabulary — teach it your jargon
• Eight export formats — TXT, Markdown, SRT, VTT, DOCX, PDF, JSON and DAPT
• A public automation CLI for scripted workflows

That combination matters. Most free Mac dictation tools do one of those three jobs. Covering offline meeting transcription on Mac and free local dictation on Mac in the same app is a real differentiator, and it is the main reason the project has traction.

Development is fast. The GitHub releases page shows v0.8.0 published on 9 September 2026, then v0.8.1 on 14 September, v0.8.2 and v0.8.3 on 16 September, and v0.8.4 on 17 September — five releases in nine days. The repository had 651 stars when checked on 17 September 2026.

What does MacParakeet cost, and what does GPL-3.0 give you?

It is free, with no paid tier at all. The official site publishes a zero price alongside the line “Everything included. No limits,” and the repository states the licence plainly: “GPL-3.0. Free software.”

For an open source Mac dictation app, that licence carries three practical consequences:

1. You can read the code. Claims about local processing are inspectable rather than taken on trust — a meaningful difference for anyone who has to justify a tool to a security reviewer.
2. You can modify and redistribute it, provided your derivative is also released under GPL-3.0.
3. Support is community-based. Bugs go to GitHub issues. There is no commercial support desk and no service-level commitment.

That third point is the honest cost of free software, not a defect. If you need someone contractually obliged to answer when dictation breaks the morning of a deadline, a GPL-3.0 project is the wrong shape of thing — regardless of how good the code is.

Which languages does MacParakeet actually handle?

Fewer than the headline number suggests, depending on which language you speak. The official site advertises 98 languages, but that figure spans several selectable engines rather than a single model.

The default local engine is Parakeet v3. NVIDIA’s model card for parakeet-tdt-0.6b-v3 — a 600-million-parameter model released under CC-BY-4.0 — describes it as “expanding language support from English to 25 European languages”, and the repository notes that v3 auto-detects among them.

The gap shows up outside Europe. The repository is candid about it: Parakeet “fails CJK/Korean here by romanizing output”, and locale-aware first-run setup selects WhisperKit instead when the Mac prefers Korean, Japanese, Chinese or Cantonese. Alternative engines — Parakeet v2, a unified English-only punctuated variant, Nemotron Beta, Cohere Transcribe and WhisperKit — remain selectable.

On accuracy, the underlying model is public and therefore checkable. NVIDIA reports an average word error rate of 6.34% on the Hugging Face Open ASR Leaderboard, an English-only evaluation across eight datasets (AMI, Earnings-22, GigaSpeech, LibriSpeech test-clean, LibriSpeech test-other, SPGI Speech, TEDLIUM-v3 and VoxPopuli). That figure says nothing about noisy rooms, accented speech or the other 24 languages, and no published benchmark compares MacParakeet against any commercial Mac dictation app.

The practical takeaway: if you dictate in a Western European language, the default engine is well matched to your use. If you work in Japanese, Korean or Chinese, expect to change engines during setup.

Is MacParakeet fully offline and private?

Speech recognition is local; the app as a whole reaches the network for several other things. The repository states that “all speech recognition runs locally” and that “your audio never leaves your Mac”, with “core capture and local transcription work offline after model setup”.

The exceptions are documented rather than hidden:

• First-run model download — the speech models have to arrive from somewhere
• Sparkle update checks — the standard macOS update framework
• Optional LLM features — if you configure one, it sends text context to the provider you choose
• Telemetry and crash reports — sent to the developer’s self-hosted endpoint by default, opt-out rather than opt-in
• The Discover feed — requests a JSON file from macparakeet.com at every launch by default, even with telemetry off and the Discover page unopened

The LLM item is the one to watch most closely in a regulated workflow: text sent to a third-party provider is an outbound transfer like any other, and it is opt-in. Telemetry and the Discover feed are not opt-in, and the repository is explicit that neither is a single global no-network toggle. If you handle privileged material, the correct posture is to turn off LLM features, telemetry and Discover, and document that decision — the same discipline that applies when dictating reports and meeting minutes offline.

Where does MacParakeet stop?

Three hard limits, none of which are bugs:

• Apple Silicon only. M1 or newer. Intel Macs are excluded entirely.
• macOS 14.2 or later. The v0.8.1 release notes repeat it: “MacParakeet requires Apple Silicon and macOS 14.2 or later.”
• No Windows build. None exists and none is announced anywhere in the repository.

There is also a softer limit worth naming: five releases in nine days is the cadence of an actively developed 0.x project. That is a good sign for momentum and a mixed one for stability. If your dictation tool needs to behave identically every morning for a year, a pre-1.0 release train is something to weigh rather than ignore.

Ready to compare against a commercial option? You can download the 15-day trial with no card required, which is the only reliable way to test dictation against your own voice and vocabulary.

MacParakeet vs Weesper Neon Flow: which fits your setup?

The honest split is platform coverage, not quality. Both run speech recognition locally, both use open speech models, and neither has published a head-to-head accuracy benchmark against the other. If you are still narrowing down the field beyond these two, our guide to choosing voice dictation software walks through the criteria that matter most.

Feature	MacParakeet 0.8.4	Weesper Neon Flow
Platforms	macOS 14.2+, Apple Silicon only	macOS 13+ (Apple Silicon and Intel with AVX2), Windows 10/11
Price	Free	€5/mo, €45/yr, or €99 one-time
Free trial	Not applicable — free	15 days
Local speech recognition	✅	✅
Default engine	Parakeet v3	Whisper (local)
Languages	98 advertised across engines; 25 European on the default engine	55+
System-wide dictation	✅	✅
Custom vocabulary	✅	✅
Meeting recording	✅ system audio + microphone	❌
File and URL transcription	✅ built in	Separate macOS app (Weesper Transcribe)
Source code available	✅ GPL-3.0	❌ proprietary
Commercial support	❌ community only	✅

Reading that table honestly: MacParakeet wins on price, on source availability and on breadth of local Mac workflow. Weesper Neon Flow’s defensible difference is narrower and specific — one consistent dictation workflow across macOS and Windows, with the same hotkey, the same custom vocabulary and the same language set on both machines.

If you only ever use a Mac, that difference is worth nothing to you.

Who should choose which?

Choose MacParakeet if:

• You work exclusively on an Apple Silicon Mac running macOS 14.2 or later
• You record and transcribe meetings, not just dictate
• Inspectable source code matters to you or to your security reviewer
• You are comfortable with GitHub issues as your support channel

Choose a paid cross-platform tool if:

• You dictate on both macOS and Windows and want one workflow, not two
• You still use an Intel Mac
• You need commercial support with a response commitment
• You want a stable release train rather than a fast-moving 0.x

For the broader field, our roundup of free offline dictation apps covers the category, and the MacWhisper and Superwhisper comparison looks at the two best-known paid Mac alternatives.

The verdict

MacParakeet is one of the more complete free local dictation tools available for the Mac in 2026. Dictation, file and URL transcription, and meeting recording with speaker labels in a single GPL-3.0 app, with no quota and no account, is a genuinely strong package — and the code is there to read.

Its boundary is equally clear: Apple Silicon, macOS 14.2 or later, no Windows, and community support. If you live inside that boundary, the free option is a serious one and the paid alternatives have to earn their keep.

If you cross that boundary daily — a Mac at the desk, a Windows machine elsewhere — one tool that behaves the same on both is the thing you are actually buying. Try Weesper Neon Flow free to see whether that consistency is worth it for your workflow, or browse the Help Center for setup guides on either platform.