OpenWhispr

Dictate and translate into the app you are already using through global hotkeys

The official README centers the product on a global hotkey: speak and the transcription is automatically pasted at the current cursor. A dedicated translation hotkey can dictate in one language and paste the result in another, bringing voice input into email, chat, IDEs, and other desktop applications without moving to a dedicated editor.

Sources: [1]

Choose local Whisper or Parakeet-style models, or cloud providers, for each workload

OpenWhispr documents local speech-to-text options including Orukeet, Whisper, NVIDIA Parakeet, and Cohere Transcribe alongside cloud processing. When a local model is selected, the project states that audio stays on the device, and GPU-accelerated local Whisper can use Metal, CUDA, or Vulkan. AI reasoning can likewise use local models or cloud providers such as GPT, Claude, Gemini, Groq, and OpenRouter.

Sources: [1]

Bring meeting detection, speaker diarization, and searchable notes into the same desktop app

The README describes automatic detection for Zoom, Teams, and FaceTime meetings, live speaker diarization, voice fingerprinting, and Google, Microsoft, or Apple Calendar integration. Existing audio and video can also be imported, while notes add folders, semantic search, AI actions, and cloud sync.

Sources: [1]

Expose notes and transcriptions to external automation through a public API and MCP

The Voice Assistant hotkey sends speech directly to an AI assistant, can edit highlighted text, and can optionally include a screenshot of the current screen as context. OpenWhispr also publishes an API and MCP server for programmatic note and transcription management or for connecting external AI assistants.

Sources: [1]

Useful when voice input should become a desktop workflow rather than a one-off transcription tool

OpenWhispr fits users who want dictation across many applications, teams or individuals consolidating meeting transcription and speaker identification, and people who want local STT or LLM options instead of a cloud-only path. Its API and MCP surface also make it relevant for developers who want notes and transcriptions to feed automation or agent workflows.

Local and cloud modes have different data boundaries, with platform-specific feature gaps and a heavier source setup

A privacy-first desktop app is not automatically a fully local workflow. Cloud transcription or hosted AI providers process audio, text, or context outside the device. A fully local setup should verify transcription, AI reasoning, speaker diarization, and semantic search individually, and users should decide whether the Voice Assistant needs screenshot context. Team spaces, web sharing, and cloud sync are also separate online surfaces.

On Intel Macs, the README says live speaker identification and voice fingerprinting are unavailable because ONNX Runtime stopped shipping macOS x86_64 binaries, and notes search falls back from semantic to keyword matching. Running from source requires Node.js 24+, while development startup downloads native helpers and model/runtime assets. The source is MIT licensed.

Sources: [1][2][3]