OpenWhispr

OpenWhispr turns spoken input into text at the cursor and supports meeting transcription, notes and optional AI assistance. It is intended for people who dictate documents or messages, accessibility workflows, and teams capturing meetings on macOS, Windows or Linux.

Voice and note workflows

• Start dictation with a global hotkey and paste the resulting text into another app.
• Choose local speech models or supported cloud transcription providers.
• Translate dictated text or apply optional AI cleanup and assistant commands.
• Record meetings, import supported audio or video files, and generate summaries.
• Organize notes and use optional account-based sync, sharing and team spaces.
• Configure dictionaries, snippets, output preferences and model/provider settings.

macOS installation and hardware

Download the signed DMG for Apple Silicon or Intel, drag the app to Applications and complete onboarding, or use brew install --cask openwhispr. The exact 1.10.2 arm64 DMG is 333,938,591 bytes, verified by HEAD against the official release asset. The macOS guide requires macOS 12 Monterey or later; capturing the other side of a meeting through system audio requires macOS 14.2 or later. A microphone is required. The official requirements page does not specify a numeric minimum RAM.

Local transcription and reasoning models require additional downloads, storage and compute. Small Whisper models occupy hundreds of megabytes, larger models several gigabytes, and local reasoning models can be larger again. The guide recommends recent Apple Silicon or suitable GPU hardware for larger models; model choice matters more than treating installer size as the complete requirement. Metal acceleration is available on Apple Silicon. Some components have newer OS requirements than the base app. The packaged app does not require a user-installed Node.js toolchain; the documented source-development workflow requires Node.js 24 or later. The cask declares no named conflict.

Permissions and network

Microphone permission allows recording. Accessibility permission enables automatic pasting; without it, copy and paste manually from the clipboard. System-audio capture uses macOS's recording authorization and is optional for meetings. Optional screen context for the assistant needs Screen Recording authorization and captures a screenshot for that command; leave it off when not needed. Grant calendar access only for a calendar integration you choose. The guide's update description may lag the 1.10.0 release, which introduced background-download automatic updates; review the actual update setting rather than relying on older instructions.

Local processing can work offline after the selected models are downloaded. Configure both transcription and any cleanup or reasoning step locally if the full workflow must stay on-device. Cloud providers, account-based sync, calendar integrations and shared notes require internet access and have their own privacy, retention and account terms. Bring-your-own-key providers require their credentials; hosted OpenWhispr Cloud and enterprise deployments have plan limits or paid services. The desktop source is MIT-licensed and free for personal and commercial use, which does not imply unlimited free cloud processing.

Privacy and accuracy

Obtain appropriate consent before recording meetings. Cloud transcription sends audio, and cloud reasoning can send transcripts or screen context to the configured service. Local settings alone do not establish that every optional integration stays offline. Protect API keys and local notes, review sharing visibility, and back up important recordings before clearing app data. AI summaries, translation and cleanup can omit or change meaning: retain the original transcript and verify decisions, numbers and action items before sharing. Current 1.10.2 notes specifically address summary timeouts, empty recordings and OpenRouter screenshot capability lookup.

Sources: Official project, macOS guide, System requirements, Official releases.