Open Source
Explore the latest AI open-source projects from GitHub and HuggingFace.
Explore the latest AI open-source projects from GitHub and HuggingFace.
Phorminx is a local-first Windows desktop application for dictation and meeting transcription, released under the MIT license by impossibleG as alpha software targeting Windows 10 and 11 on x64. Speech recognition runs entirely on-device: an Instant mode uses the Vosk toolkit for immediate resident transcription, while an Accurate mode runs Whisper with optional Vulkan GPU acceleration; both begin transcribing while the user is still speaking, drawing from a bounded rolling audio buffer that production capture caps at 45 seconds of 16 kHz mono audio held in memory rather than written to disk. A default Ctrl+Alt+Space shortcut opens a launcher for starting and stopping dictation, and a separate Meetings workspace captures microphone or computer audio, keeps live transcripts, and offers a floating companion panel that can answer a question using the transcript context captured so far without interrupting capture. Retained text and settings are stored locally under %LOCALAPPDATA%\Phorminx in a SQLite database, with configurable history retention windows of 24 hours, 7 days, 30 days, or indefinite, plus a personal lexicon and per-application formatting profiles keyed to executable names; the persistence layer does not record window titles. Optional formatting can run through a local Ollama model, and a separate transport crate adds optional remote assistant support through OpenAI's Responses API (explicit model identifiers only) and Anthropic's Messages API, with request cancellation that closes the underlying connection. Speech models and the Vosk runtime are downloaded separately and are not bundled in the source repository, and installers, compiled binaries, and model weights carry their own licensing terms distinct from the project's MIT-licensed source. The codebase is organized as a Rust workspace of about a dozen crates covering the desktop app, audio capture, session and persistence logic, the Vosk and Whisper recognition backends, Windows integration, and the Ollama and remote-assistant transports, built with Visual Studio's C++ tooling, CMake, and the Vulkan SDK for GPU-enabled builds, and packaged into a Windows installer with Inno Setup.
ggml-org
Pure C/C++ port of OpenAI Whisper for edge deployment
CJ Pais
A free, open-source, cross-platform speech-to-text app that transcribes your voice entirely offline — press a shortcut, speak, and have the text pasted into any app.