A free, open source, and extensible speech-to-text application that works completely offline.
26.1k
Stars
2.2k
Forks
163
Open issues
30
Contributors
AI Analysis
Handy is a free, open-source desktop application for offline speech-to-text transcription using local models (Whisper, Parakeet V3) with privacy-first design. It serves accessibility and power users who need transcription without cloud dependency, particularly benefiting people with motor disabilities, privacy-conscious users, and those in environments without reliable internet. It is not intended for users requiring cloud-based advanced features, real-time collaborative transcription, or hig...
Inferred from signals mentioned in the README (tests, CI, type safety) — not a review of the actual code.
AI's overall editorial judgment — not an average of the bars above, can weigh other factors too.
Handy brings fully offline, open-source speech-to-text to any desktop text field via a global hotkey
Handy is a cross-platform desktop app (Windows, macOS, Linux) that transcribes speech locally using Whisper or Parakeet models and pastes the result into any active text field. It targets privacy-conscious users, developers, and accessibility-focused individuals who want dictation without cloud dependency. Built with Tauri (Rust backend, React frontend), it supports GPU acceleration, VAD filtering, configurable shortcuts, and CLI remote control. With ~24,700 stars gained largely since early 2025, it has attracted meaningful community attention, a Raycast extension, Homebrew and winget packaging, and an active Discord.
Created in February 2025, Handy is a young project that appears to have emerged from the wave of interest in local Whisper-based tooling following OpenAI's open-source model releases. It explicitly positions itself as 'the most forkable' speech-to-text tool, prioritizing extensibility over feature maximalism.
The project accumulated ~24,700 stars in roughly 16 months, with 922 stars in the last 7 days alone — suggesting sustained organic discovery, likely driven by privacy and offline-AI trends, social sharing among developer communities, and the growing desire for local AI tooling. macOS Homebrew cask and winget availability lowered the installation barrier significantly.
Third-party Raycast extension exists with a separate maintainer, indicating at least some real-world usage beyond casual evaluation. Homebrew cask and winget entries exist, though the README notes these are not maintained by the core team. Discord server is active. Concrete user counts and deployment-at-scale data are not publicly documented.
Appears to be a Tauri v2 application combining a Rust backend for audio capture (cpal), VAD (vad-rs/Silero), ML inference (whisper-rs, transcribe-rs/Parakeet), and system events (rdev), with a React/TypeScript/Tailwind frontend for settings UI. Likely single-binary distribution with bundled model downloads. The CLI remote-control via single-instance plugin suggests a clean IPC design.
Not documented in README
Last push was on the same day as this analysis (2026-06-24), indicating active, current development. Known issues are transparently documented in the README with 'Help Wanted' labels. The presence of a BUILD.md, debug mode, CLI flags, and multiple model options suggests sustained iterative development over 16 months.
ADOPT IF: you need privacy-first, fully offline dictation across any desktop OS, are comfortable with minor rough edges in an actively developed tool, or want a forkable base for custom transcription workflows. AVOID IF: you need rock-solid stability on all configurations right now — known Whisper crashes on some Windows/Linux setups are a real issue — or require enterprise-grade support and polish. MONITOR IF: you are on Linux with Wayland (support is limited and requires third-party tools) or are evaluating it for accessibility-critical workflows where reliability is non-negotiable.
Independent dimensions
Mainstream potential
6/10
Technical importance
7/10
Adoption evidence
4/10
- Whisper model crashes on certain Windows and Linux configurations are a documented, unresolved issue that may affect some users in unpredictable ways.
- Wayland support on Linux is incomplete and depends on external tools (wtype, dotool), limiting out-of-box usability for a growing segment of Linux desktop users.
- The project is ~16 months old and architecturally young; breaking changes or significant refactors are plausible as it matures.
- Homebrew and winget packages are community-maintained rather than by core developers, creating a potential lag in updates for non-source users.
- Dependency on upstream Whisper and Parakeet model ecosystems means that breaking changes in those libraries could require significant maintenance effort.
Likely to consolidate as a go-to open-source desktop dictation tool for privacy-conscious users; may expand model support and OS integration depth, but faces increasing competition as the local-AI desktop tooling space fills in.
Newsletter
Get analyses like this every Monday
Free weekly digest of the most interesting open-source discoveries.
Languages
Information
- Website
- https://handy.computer
- Language
- Rust
- License
- MIT
- Last updated
- 1d ago
- Created
- 17mo ago
- Analyzed with
- anthropic/claude-haiku-4-5
Stars over time
Contributors over time
Top 100 contributors only — repos with more will plateau at 100.
Open issues
[BUG] v0.9.0 takes too long on all models
[BUG] App segfaults (SIGSEGV) on launch enumerating audio devices on macOS 26 (Tahoe)
[BUG] Crash everytime transcribe is fired
[BUG] deb package: bundled libs (onnxruntime/ggml/transcribe) installed as duplicate real files instead of symlinks, triggers ldconfig warnings
[BUG] In portable mode, models are not saved to the application folder
Top contributors
Recent releases
Similar repos
OpenWhispr/openwhispr
OpenWhispr is a privacy-first, open-source voice-to-text application for macOS,...
| Repository | Stars | Week Δ | Language | Score | Updated |
|---|---|---|---|---|---|
|
cjpais/Handy
this
|
26.1k | +715 | Rust | 8/10 | 1d ago |
|
OpenWhispr/openwhispr
|
4.4k | — | TypeScript | 7/10 | 8h ago |
 thewh1teagle/vibe
|
6.7k | — | TypeScript | 8/10 | 6d ago |
 altic-dev/FluidVoice
|
7k | — | Swift | 7/10 | 13h ago |
 goodroot/hyprwhspr
|
1.1k | — | Python | 7/10 | 2d ago |
 Zackriya-Solutions/meetily
|
22.5k | — | Rust | 7/10 | 1mo ago |
VoiceInk is macOS-only and targets a similar privacy-first dictation niche. Handy is cross-platform and explicitly more extensible/forkable, but VoiceInk may offer a more polished native macOS experience.
Vibe focuses on transcribing audio/video files rather than live system-wide dictation. Different primary use case — Handy is for real-time typing assistance, vibe is for post-processing recordings.
Meetily targets meeting transcription and summarization workflows, not general-purpose system dictation. The two tools serve overlapping audiences but distinct workflows.
OpenWhispr appears to serve a similar local Whisper transcription niche. Handy has significantly more stars and offers broader model support (Parakeet in addition to Whisper), suggesting it may have broader adoption or better discoverability.
OS-native dictation works well but requires cloud connectivity (or limited local models) and is not extensible or open source. Handy trades some polish for full offline operation and hackability.
