OpenWispr

Open-source voice-to-text, honestly reviewed. Local models, no subscriptions, no cloud audio uploads unless you want them.

TL;DR

• What it is: Open-source (MIT) voice dictation app — think Wispr Flow, but your audio runs through a Whisper or NVIDIA Parakeet model on your own hardware, not someone else’s server [1][2].
• Who it’s for: Developers, writers, and founders who dictate constantly and are done paying $10–15/month to cloud-based tools that quietly upload every word they speak [1][3].
• Cost savings: Wispr Flow costs $144/year (or $15/month). OpenWispr’s software is free; local models cost $0/month to run. Even the Pro cloud tier is competitive for unlimited words [website].
• Key strength: Truly local — audio never leaves your machine when using Whisper or Parakeet locally. Hold a key, speak, release, text appears wherever your cursor is. Global hotkey works in VS Code, Slack, browser fields, terminal [1].
• Key weakness: It’s a young project (1,972 GitHub stars) and the AI cleanup layer is less refined than Wispr Flow’s. You’re trading polish for privacy and cost [4].

---

What is OpenWispr

OpenWispr is a cross-platform voice dictation and productivity app. Press a hotkey anywhere on your computer, speak, release — your transcribed words appear at cursor position in whatever app is focused. VS Code, Slack, Gmail, a terminal prompt, a browser text field. It works wherever you can type [1].

The founder built the first version after looking at the macOS dictation market and finding a depressing range of options: Apple Dictation (free but butchers technical terms), Wispr Flow ($15/month, cloud-only audio upload), Superwhisper ($249 lifetime), and MacWhisper ($80, but only works inside its own app). His conclusion was blunt: “Charging $10/month or $249 for what amounts to a thin wrapper around an open-source model felt wrong to me” [1]. The entire stack — Whisper, Metal GPU acceleration on Apple Silicon, Swift, Homebrew — was already free.

What started as a lean macOS hotkey-to-cursor tool has since grown into a fuller product. As of the current version, OpenWispr includes dictation across 100+ languages, an AI agent overlay for voice commands, meeting transcription with Google Calendar integration, a notes system with full-text search, a custom dictionary with auto-learning, and both local and cloud (BYOK) transcription paths [README]. It runs on macOS, Windows, and Linux.

The project is MIT-licensed. The code is public on GitHub for anyone to audit, fork, or self-host [1][README].

---

Why people choose it

The through-line across every review is the same: cloud dictation tools ask you to trust them with audio data that’s genuinely sensitive, and OpenWispr is the alternative where you don’t have to [1][3][4].

Versus Wispr Flow. This is the primary comparison. Wispr Flow captures screenshots of your active window every few seconds and sends them to cloud servers alongside your audio. It costs $144/year, has a free tier capped at 2,000 words/week, and has no local processing option. One Voibe article [3] clocks it at 800MB RAM and 8% CPU at idle, with startup times of 8–10 seconds. For a tool designed for quick dictation, that startup cost is real friction. The price gap is obvious: Wispr Flow’s annual plan costs more than a cheap VPS for a year, and at the end of that year you have no software, no data, nothing [1][3].

One Threads post [2] frames it simply: OpenWispr is the open-source alternative to Wispr Flow, Superwhisper, Willow, Aqua Voice, and Monologue. Multiple testimonials on the OpenWispr homepage echo this — one user notes: “I was using Aqua Voice before, but this is OSS with the same features and feel — plus you can choose between API-based pay-as-you-go or downloading a local ASR model” [website testimonials].

Versus Superwhisper. Superwhisper charges $249 lifetime and is Mac-only. OpenWispr is cross-platform and costs $0 for local use [1][3]. For users who need to work across operating systems or don’t want to pay $249 upfront for a productivity tool they might try and abandon, this matters.

On the privacy question specifically. The creator was explicit about the design intent: “Voice data is intimate. It’s not like search queries or browsing history — it’s literally you, speaking, captured in audio. When a dictation app sends that to a server, you’re trusting a company with a recording of your voice, the content of what you said, and the context of when and where you said it.” [1] The local processing path — Whisper or NVIDIA Parakeet running on your GPU — makes no network requests. The audio file exists for milliseconds, gets processed, and is deleted [1][website].

On performance. The creator found Whisper accuracy better than every commercial dictation tool they’d used: “It handles code terms like ‘useState,’ ‘kubectl,’ and ‘JWT’ without breaking a sweat.” On Apple M1, the base model transcribes in real-time or faster with Metal GPU acceleration [1]. One Twitter/X testimonial: “Whisper Base local is near instant in OpenWhispr, and I found it better than Parakeet. Parakeet is near instant, too.” [website testimonials]

---

Features

Core dictation:

• Global hotkey (default: backtick) to start/stop dictation from anywhere — pastes at cursor [README][1]
• Automatic pasting with toggle to disable [README]
• NVIDIA Parakeet local transcription via sherpa-onnx (25 languages) [README][website]
• Whisper model support: tiny (75MB), base (142MB), small (466MB), medium (1.5GB), turbo (1.6GB) — you choose the speed/accuracy tradeoff [website]
• 100+ languages with auto-detection [website]
• Custom dictionary with auto-learn from corrections [README][website]
• SQLite transcription history with audio retention and retry [README]
• Auto-pauses media (Spotify, Apple Music) during dictation and resumes after [README]
• Works offline — no wifi needed when running local models [website]

AI agent mode:

• Glassmorphism overlay with real-time AI streaming and conversation history [README]
• Custom agent name — say “Hey [AgentName]” to give voice commands [README]
• Multi-provider AI: OpenAI (GPT-4.1, GPT-5, o-series), Anthropic (Claude Sonnet 4.6, Opus 4.6), Google Gemini, Groq, and local models via llama.cpp [README]
• BYOK — bring your own API keys for any supported provider [website]
• AI actions on notes: apply customizable processing templates by voice command [README]

Meeting transcription:

• Google Calendar integration with auto-detection of upcoming meetings [README]
• Live transcription via OpenAI Realtime API with WebSocket streaming [README]
• Auto-detects Zoom, Teams, FaceTime [README]
• Dedicated meeting hotkey separate from dictation hotkey [README]

Notes:

• Notes with folder organization, drag-and-drop, audio upload [README]
• FTS5-powered full-text search with Cmd+K command palette [README]
• Cloud sync option with local-first storage and semantic search [README]

Cloud option:

• OpenWispr Cloud sign-in — no API keys needed, free and Pro tiers [README]
• Free tier: 2,000 words/week [README]
• Pro: unlimited words [README]
• 7-day free trial [README]
• Google OAuth and email/password sign-in [README]

---

Pricing: SaaS vs self-hosted math

OpenWispr Free tier:

• 2,000 words/week on cloud transcription
• All local model features: unlimited
• No API keys required for basic use [website][README]

OpenWispr Pro:

• Unlimited cloud words
• 7-day free trial
• Pricing not published on the website at time of review — contact or sign up to see current rate [website]

BYOK / local:

• $0 software cost
• Pay only for API calls if using OpenAI, Anthropic, Groq directly
• Local Whisper or Parakeet: no per-word cost, no subscription [website][README]

Wispr Flow for comparison [3]:

• Free: 2,000 words/week (cloud-only)
• Pro: $15/month or $144/year
• No local option, audio goes to cloud regardless

Superwhisper: $249 lifetime, Mac-only [1][3]

MacWhisper: $80 lifetime, only works in its own app [1]

Concrete math for a developer who dictates constantly:

Say you dictate 30 minutes a day — roughly 5,000–8,000 words at natural speech pace. On Wispr Flow’s free tier you blow past 2,000 words within a single work day. Pro at $144/year is your only option if you want cloud. Self-hosting OpenWispr with the small Whisper model (466MB) on a machine you already own: $0/month, unlimited words, forever [1][website].

If you don’t want to manage local models and prefer their cloud: wait for published Pro pricing. If the Pro tier comes in under $60/year (likely, given competitive positioning), it’s still a 60%+ discount from Wispr Flow.

The honest caveat: local model setup requires comfort with downloading 142MB–1.6GB model files and understanding the speed/accuracy tradeoff. If you pick a model that’s too slow for your hardware, you’ll notice latency. The creator reports M1 handles the base model in real-time [1].

---

Deployment reality check

For local use on macOS, the install path is Homebrew [1]. The creator describes it as a one-command install that runs as a background service with a menu bar icon. On Windows and Linux, the cross-platform Electron-based distribution is available via the GitHub releases page [README].

What you actually need for pure local use:

• macOS, Windows, or Linux machine with a microphone
• Homebrew (macOS) or downloaded binary (all platforms)
• 150MB–1.6GB free disk space for whichever Whisper model you choose
• Accessibility permission (to type at cursor) and microphone permission — these are the macOS permission prompts that took the creator time to get right [1]

What you need for AI agent mode:

• An API key for OpenAI, Anthropic, Groq, or Google (BYOK) — or a local llama.cpp setup for fully offline AI [README]

What can go sideways:

• The macOS Accessibility permission prompt is genuinely the hardest part. The creator flagged it explicitly: “The hardest part wasn’t the AI. It was the macOS permissions.” Expect to spend a few minutes granting both Accessibility and Microphone access and possibly restarting the app [1].
• The Voibe comparison article [3] benchmarks local transcription on modern hardware as sub-second, but your mileage varies significantly based on which model you pick and your GPU. Whisper large (1.6GB) on a CPU will be noticeably slower than base (142MB).
• Meeting transcription requires Google Calendar OAuth setup and uses OpenAI Realtime API — this is cloud-dependent and requires an OpenAI API key [README].
• This is a young project at 1,972 GitHub stars. The meeting transcription and notes features are newer than the core dictation path, and polish in those areas is still catching up [README].

Realistic setup time for a technical user: 10–20 minutes including permissions, model download, and hotkey customization [1][4]. Non-technical users on macOS: 30–60 minutes with some Google searching about Accessibility permissions.

---

Pros and cons

Pros

• Actually MIT-licensed. Every line of code is public. You can verify the privacy claims yourself — there’s no trust-me-we’re-different marketing required [1][website].
• Truly local when you want it. The local Whisper + Parakeet path makes zero network requests. Audio file exists for milliseconds then is deleted [1][website]. This isn’t an add-on — it’s the primary path.
• Works everywhere. Global hotkey pastes at cursor in any app: VS Code, Slack, terminal, browser, email client [1][website]. This is the core feature and it works as advertised.
• Whisper accuracy is genuinely good. Technical jargon, code terms, proper punctuation — multiple users confirm it handles developer vocabulary better than Apple Dictation and on par with cloud commercial alternatives [1][website testimonials].
• No per-word pricing anxiety. Local use is unlimited. No subscription ticking in the background while you dictate a long document [1][3].
• BYOK for cloud models. If you want cloud AI quality, you bring your own API key and pay OpenAI/Anthropic directly — no markup, no vendor lock-in [website][README].
• Cross-platform. macOS, Windows, Linux — not the Mac-only constraint of Superwhisper and MacWhisper [README][2].
• 100+ languages. Auto-detection included [website].

Cons

• AI cleanup layer is less refined than Wispr Flow. Wispr Flow rewrites your dictation — strips filler words, fixes grammar, adapts to your tone. OpenWispr transcribes what you say and offers optional AI cleanup, but it hasn’t had years of refinement behind it [4]. If you speak in full polished sentences, this gap is small. If you speak messily, it’s noticeable.
• Young project. 1,972 GitHub stars. Meeting transcription and notes features are newer additions, not battle-tested features. Some rough edges exist [README]. This isn’t a company with an enterprise support line.
• Pro pricing not published. The website doesn’t show the Pro tier price. You have to sign up to find out. Minor but annoying if you’re comparison shopping [website].
• Meeting transcription requires cloud. The live meeting transcription feature uses OpenAI Realtime API — not local [README]. If you wanted fully offline meeting transcription, this isn’t it.
• Accessibility permission friction on macOS. The macOS Accessibility permission dance is real. It’s a one-time setup but it trips up non-technical users [1].
• No standalone server/API mode. Unlike Whisper WebUI [5] which exposes a web interface for batch transcription from any device, OpenWispr is a desktop app. If you want a transcription endpoint your whole team hits, you need different tooling.
• Model download on first run. Whisper base is 142MB, turbo is 1.6GB. First-run download isn’t fast on slower connections [website].

---

Who should use this / who shouldn’t

Use OpenWispr if:

• You’re a developer or founder who dictates frequently — emails, prompts, documentation, Slack messages — and paying $10–15/month for Wispr Flow bothers you on principle.
• Privacy is a real constraint, not just a preference. Working with client-confidential information, proprietary code, or legal documents that shouldn’t leave your machine [1][4].
• You’re on Linux or Windows and the Mac-only alternatives (Superwhisper, MacWhisper) aren’t options [README][2].
• You want global-hotkey dictation that works everywhere, not a dedicated transcription window.
• You’re comfortable spending 15 minutes on setup — Homebrew install, permissions, model download — once.

Skip it (stay on Wispr Flow) if:

• You want the best AI cleanup layer available. Wispr Flow’s rewriting is more refined, and if you speak informally and want polished output automatically, that matters [4].
• You don’t want to think about model selection, RAM usage, or Accessibility permissions. Wispr Flow works on first install with no configuration.
• Your employer’s IT policy requires cloud-certified tools with a support contract.

Skip it (try Superwhisper) if:

• You’re Mac-only, want context-aware formatting modes (email mode, code mode, etc.), and don’t mind paying once for a more polished experience [3].

Skip it (use Whisper WebUI) if:

• You need a web-accessible transcription service for your whole team — batch file transcription, subtitle generation, YouTube video transcription. Whisper WebUI is a server with a web interface; OpenWispr is a desktop app [5].

---

Alternatives worth considering

• Wispr Flow — the incumbent cloud dictation tool. Better AI cleanup, iOS support, easier setup, $144/year, no local option, sends audio + screenshots to cloud [1][3][4].
• Superwhisper — Mac-only, $249 lifetime, context-aware formatting modes, offline capable. More polished than OpenWispr currently, but Mac-only and costs money [1][3].
• MacWhisper — $80 lifetime, Mac-only, only works inside its own app. Not suitable for global hotkey dictation [1].
• VoiceInk — free, open-source, Mac-focused, 100+ languages. Mentioned in the Wispr alternatives roundup [3]. Less widely reviewed than OpenWispr.
• Voibe — 100% offline, sub-second startup, $44.10/year or $99 lifetime. Privacy-focused commercial option if you want local processing with commercial support [3].
• Whisper WebUI — Docker-based web interface for Whisper. Best for team/server transcription use cases, not desktop dictation [5].
• Apple Dictation — free, built-in, works offline on modern macOS. Falls apart on technical vocabulary and doesn’t paste globally as reliably [1].
• Diction — homelab-focused alternative built by a developer who also left Wispr Flow. Go gateway + faster-whisper server in Docker Compose, iOS keyboard extension. Self-hosted pipeline for phone use [4].

For a non-technical founder comparison shopping between dictation tools: the realistic shortlist is OpenWispr vs Wispr Flow. OpenWispr if local privacy and zero recurring cost matter. Wispr Flow if polish and iOS support matter more than data sovereignty.

---

Bottom line

OpenWispr is the right answer to a legitimate question: why should voice dictation cost $144/year and route your audio through someone else’s cloud when OpenAI Whisper runs locally on the hardware you already own? The creator built the first version in frustration, the project has grown into something substantially more capable, and the MIT license means the answer to “can I trust their privacy claims?” is “read the code.” That’s a better answer than any privacy policy.

The trade-offs are real: the AI cleanup layer isn’t as refined as Wispr Flow’s, the project is young and some features are still rough, and setup takes 15 minutes versus Wispr Flow’s zero. For developers, technical writers, and founders handling sensitive content who dictate regularly — the $0/month math and the local-first architecture are an obvious call. For anyone who wants to install and forget and doesn’t care about audio privacy, Wispr Flow is a better product right now.

If you need help with the deployment setup, that’s exactly the kind of one-time infrastructure work upready.dev handles for clients.

---

Sources

1. Ammon Taylor, Medium — “I Built a Free, Open-Source Voice Dictation App for Mac Because the Alternatives Are Absurd” (Feb 16, 2026). https://medium.com/@ammonx9/i-built-a-free-open-source-voice-dictation-app-for-mac-because-the-alternatives-are-absurd-43ab9ca74ae9
2. Open Source Alternatives, Threads — “OpenWispr ⭐️ 429 — Turn voice into clean, polished text 3x faster than typing” (Feb 2025). https://www.threads.com/@opensourcealternatives/post/DRx0pxTCo7G/open-wispr-turn-voice-into-clean-polished-text-x-faster-than-typing-privacy
3. Voibe Blog — “9 Best Wispr Flow Alternatives in 2026: Privacy, Pricing, and Performance”. https://www.getvoibe.com/blog/wispr-flow-alternatives/
4. omachala, DEV Community — “I Stopped Paying $15/Month for Wispr Flow. Here’s the Open-Source Replacement.”. https://dev.to/omachala/i-stopped-paying-15month-for-wispr-flow-heres-the-open-source-replacement-313i
5. noted.lol — “Whisper WebUI — The Self-Hosted AI Transcriber”. https://noted.lol/whisper-webui/

Primary sources:

• GitHub repository and README: https://github.com/herotools/open-wispr (1,972 stars, MIT license)
• Official website: https://openwispr.com
• Homepage and features: https://openwhispr.com