TheHive Voice is an open-source voice operating layer for macOS, iOS, and watchOS. The product direction combines dictation, text-to-speech, realtime conversation, and governed agent control across local, tailnet, BYOK, and cloud routes.
Important
The repository is in its foundation phase. The current desktop implementation is inherited from OpenWhispr. The additional TheHive Voice modes and Apple-platform clients are product direction, not released functionality yet.
The inherited OpenWhispr baseline already provides a substantial cross-platform desktop surface:
- global voice dictation and paste;
- local Whisper and NVIDIA Parakeet transcription;
- cloud transcription and AI-provider routes;
- meeting transcription, notes, and local semantic search;
- Electron, React 19, TypeScript, Tailwind CSS v4, and shadcn/ui.
The first TheHive Voice work builds upward from that functional baseline. It does not replace working dictation with a prototype.
- Voice-to-text for fast system-wide dictation.
- Text-to-speech with explicit local or remote routing.
- Realtime conversation with clear microphone, connection, and session state.
- Agent mode for user-selected Codex, Hermes, OpenClaw, OpenCode, or compatible endpoints without silent fallback.
- Apple continuity across macOS, iOS, and watchOS.
- Local-first and BYOK choices alongside optional managed services.
- Visible authority and stop controls whenever an agent can act on a device.
There are no TheHive Voice releases yet. Do not use upstream OpenWhispr release artifacts as proof of a TheHive Voice release.
For the inherited desktop baseline:
git clone https://github.com/TheAngryPit/TheHiveVoice.git
cd TheHiveVoice
nvm use
npm ci
npm run devNode.js 24 is required. Native resources and model downloads make the first development start materially heavier than a normal React application.
mainmirrorsOpenWhispr/openwhispr:main.productis the default TheHive Voice integration branch.- Upstreamable fixes branch from
mainand target OpenWhispr upstream. - TheHive Voice features branch from
productand target this repository. - Upstream updates reach
productthrough a reviewable reconciliation pull request, never by copying files manually.
Read the upstream and worktree workflow before starting implementation.
Use focused local checks during development:
npm test
npm run typecheck
npm run lintBroad, expensive, or CI-parity validation belongs on Crabbox when that provider is configured. A Crabbox result must report the actual provider, lease ID, command, and exit status; a queued or unavailable box is not a passing gate.
Issues and pull requests are welcome. Start with the contribution guide. Security vulnerabilities must follow the private reporting policy, not a public issue.
MIT. The fork preserves the upstream license and attribution.
TheHive Voice is built from the substantial work of the OpenWhispr maintainers and contributors. The inherited baseline also depends on:
- OpenAI Whisper
- whisper.cpp
- NVIDIA Parakeet
- sherpa-onnx
- Hugging Face
- llama.cpp
- Electron
- React
- shadcn/ui
Upstream-specific sponsors and cloud infrastructure are not presented as TheHive Voice sponsors or infrastructure unless TheHive Voice adopts them explicitly.