OpenWhispr Review: The Open Source Dictation App With Three Ways to Transcribe

Dictation software has a trust problem built into its shape. You speak private work into a microphone, and most polished apps send that audio to a server you cannot inspect, under a policy you have to take on faith. OpenWhispr starts from the opposite end. Its client code is public under the MIT licence, its local models run on your own machine, and its maker says plainly that in local mode your audio never leaves the device at all. This review is about whether that promise survives contact with the product as it ships in late 2026, cloud tier and all.

What It Is

OpenWhispr is a desktop dictation app built by Gizmo Labs Inc. and published as open source at github.com/OpenWhispr/openwhispr. The repository opened in June 2025 and has grown fast. At review time it showed about 10,550 stars, 1,373 forks and 2,537 commits, with version 1.11.0 released on October 29, 2026. The 1.11.0 notes added a shared team dictionary, assistant skills for recurring work, and a batch of new models.

The core loop is simple. Press a global hotkey, speak, press again, and the transcript lands at your cursor in whatever app holds focus: Slack, a code editor, a browser, a word processor, a terminal. Hold to record and push to toggle modes are both supported, and the hotkey itself is yours to set. Around that loop the app has grown meeting recording, notes, translation and an agent mode, which is why the maker now pitches it as an alternative to both Wispr Flow and Granola at once.

Features and Specifications

Three facts in that table do most of the work. The first is the model shelf. You are not locked to one engine: a light Whisper model answers near instantly on modest hardware, Parakeet suits fast multilingual work, and larger models trade speed for accuracy, all without a server in the loop. The second is platform reach. Linux support alone sets OpenWhispr apart in a field where the polished names cluster on Mac and Windows. The third is scope creep, in the good and risky senses at once: meetings, notes, translation and agents all live in the same resident app.

Real World Experience

This is a research review. It draws on the official site, pricing page and changelog, the public GitHub repository and README, and independent reviews published in October 2026, rather than hands on testing in our lab. Where a figure comes from the maker, it is labelled as a maker claim.

Start with speed, because dictation lives or dies on latency you feel. The maker claims about 150 words per minute with OpenWhispr against about 40 words per minute typing, which is a claim about speech itself more than about this app. Independent voices are more useful on feel. User reports quoted on the official pricing page call Whisper Base near instant locally, with Parakeet near instant as well. Published reviews agree that local models on current hardware are fast enough for daily dictation, and that cloud paths mainly help older machines or buyers chasing top accuracy.

The sharper independent finding is about defaults. A detailed October 2026 review that read the source code at a fixed commit, the README, the docs and the pricing page concluded that the app can transcribe entirely on your computer for free with no word limit, and that it is also the front end for a paid hosted service the setup screen recommends. Its advice was blunt: pick the local card during setup if privacy is the reason you came, then check your settings afterwards. That matches the structure of the pricing page itself, where the free cloud allowance is listed first and local models second. Privacy here is real, but it is privacy by configuration. You have to choose it.

Weight is the other lived cost reviewers keep naming. OpenWhispr is an Electron app, and one October review put each Mac download at about 334 MB, with separate builds for Apple silicon and Intel. As an always resident utility it sits heavier than native rivals and far heavier than a small local only tool. On a modern laptop that is a background fact. On an older machine it can be the reason the dictation feels slow even before any model loads.

The meeting side is where the scope pays off for some buyers. The app detects Zoom, Teams and FaceTime calls, records system audio without a bot joining the call, labels speakers live where the platform supports it, and files the result into searchable notes. The free plan includes 5 hours of meeting recording a month, Pro raises that to 20 hours, and Business removes the cap. One hard limit is documented in the README itself: on Intel Macs, live speaker identification and voice fingerprinting are unavailable because the ONNX Runtime stopped shipping macOS Intel binaries in version 1.24. Meetings still record and transcribe on those machines, and notes search falls back to keyword matching.

What Works

The local path is the headline and it holds up under inspection. Unlimited transcription on your own machine, in more than 100 languages, with no account barrier for the free plan and no card required, is a combination no closed rival matches. Because the code is public, the central privacy claim is checkable in a way a policy page never is. For doctors, lawyers, journalists and anyone dictating sensitive work, that audit trail plus the listed SOC 2 Type II and ISO 27001 certifications gives a security team something concrete to review.

Flexibility is the second genuine strength. Hardware varies, and so does tolerance for cloud risk. OpenWhispr lets one person run a small local model on a train with no signal, switch to a large local model at a desk, and use cloud accuracy for one hard recording, all inside one hotkey and one dictionary. The bring your own key path matters here: on any plan, your own OpenAI, Anthropic, Gemini or Groq keys carry no OpenWhispr limit, per the pricing FAQ. You pay the provider rate and keep that data relationship direct.

The third strength is pace. A repository that went from launch in mid 2025 to more than ten thousand stars and a near weekly release rhythm by autumn 2026 is not a weekend project. Version 1.11.0 shipped days before this review. Buyers of young software should still expect rough edges, and the open issue count shows them, but the direction of travel is sustained work rather than abandonment.

What Does Not

The free cloud cap is the first friction, and it is small. Two thousand words a week is roughly fifteen minutes of ordinary speech. Anyone who dictates as a primary input will hit that ceiling in days if they lean on cloud. The honest reading is that the free cloud tier is a trial path, while the free local tier is the actual free product. That is a fair deal, but the marketing leads with cloud, and new users can easily miss which tier they are really on until the meter stops them.

Polish is the second gap. Wispr Flow built its name on cleaning up rambling speech into text that reads as if you typed it on a good day, with tone that adapts to the app in focus. OpenWhispr has AI cleanup and a learning dictionary, and its agent layer can call strong models through your keys, but published comparisons still give the polish crown to the closed rival. If your dictation is mostly casual messages and prompts, the gap is minor. If you dictate finished prose all day, you will feel it in extra editing passes.

The third gap is reach and weight together. There is no Android app on the official pricing page, the iPhone and iPad app requires a paid plan, and the Electron build is heavy for a utility that never closes. A buyer whose life splits evenly between phone and desktop will find the phone half of this product thin compared with Wispr Flow, which ships on Mac, Windows, iOS and Android.

Competition

The table draws three clean lanes. Handy is the purity pick: local only, free, and focused on dictation alone, with a far larger star count than OpenWhispr in October checks of the two repositories. Wispr Flow is the polish and phone pick, at $144 a year and with every word routed through its servers. Superwhisper is the Mac centred middle path. OpenWhispr wins where a buyer wants meetings and dictation in one app, Linux on the desk, and the freedom to move between local and cloud without changing tools.

Final Verdict

Choose OpenWhispr if you will actually use the local path. Set it up local first, pick a model your hardware runs well, and you get unlimited private dictation on any desktop platform for zero dollars, backed by code you or your IT team can read. Add Pro at $80 a year only when unlimited cloud, device sync, the agent mode or the iPhone app earns its keep. At that price it still undercuts Wispr Flow Pro by $64 a year.

Choose Wispr Flow instead if phone dictation matters as much as desktop, or if finished prose polish is worth a cloud only pipeline and a higher bill. Choose Handy if you want the strictest form of the local promise, an app with no cloud path to misconfigure. OpenWhispr is not the simplest dictation app and it is not the most polished. It is the one that gives you the most control over where your voice goes, and in a category built on trust, control you can verify is the feature that counts.

This is a research review. It draws on the official OpenWhispr site, pricing page and changelog, the public GitHub repository, and independent reviews and comparisons published in October 2026 and checked November 2026. No hands on testing was performed.

Sources

• OpenWhispr: Official site
• OpenWhispr: Pricing
• OpenWhispr: Changelog, version 1.11.0
• GitHub: OpenWhispr repository
• Voibe Resources: OpenWhispr review, free local dictation with a cloud catch
• EmberType: OpenWhispr review, free, open source, cloud by default
• SpeakoFlow: Wispr Flow alternatives for Windows, checked October 17, 2026
• Product images: OpenWhispr