OpenFlowKit is an open-source, LLM-agnostic voice dictation and AI text refinement kit inspired by the best parts of Wispr Flow, Superwhisper, VoiceInk, OpenWhispr, OpenTypeless, and other modern dictation tools.
It is not a generic transcription demo. The goal is a privacy-first voice layer that works across apps, IDEs, and LLM workflows:
- speak naturally into any text field
- transcribe through local or cloud STT providers
- clean filler words, backtracks, punctuation, tone, and structure
- adapt to dictionaries, snippets, writing styles, and developer syntax
- integrate with any OpenAI-compatible LLM, Ollama, OpenRouter, Anthropic, Gemini, or self-hosted model gateway
- monetize by gating cloud convenience, team controls, usage analytics, and managed enterprise compliance
Wispr Flow is polished, but it is proprietary and cloud-heavy. Open-source alternatives cover pieces of the workflow, usually local Whisper transcription or a desktop hotkey. OpenFlowKit focuses on the missing layer: provider abstraction, app integrations, privacy controls, team policy, and a credible paid SaaS wrapper around a free core.
- Browser MVP voice capture — Web Speech API push-to-talk with real-time transcript display
- Terminal bridge — WebSocket relay for desktop injection + command mode
- Dictation pipeline — Capture → Transcribe → Refine → Inject with full latency tracking
- Enhanced refinement — Self-correction resolution, deduplication, style adaptation (concise/formal), document formatting
- Comprehensive tests — Refinement, capture, and dictation pipeline test coverage
- Strategic report — Competitor analysis, market sizing, profit margins, five moats, and 3-year financial model in
docs/REPORT_2026-06-03.md
This first repository ships:
- a typed TypeScript core for dictation requests, provider contracts, refinement, and monetisation tiers
- a working Vite/React web surface with live voice capture, terminal bridge, integrations, pricing, and architecture sections
- research notes and product architecture docs
- a build output path of
dist
Desktop capture, native paste injection, and background hotkeys should be implemented next with Tauri or Electron. The core interfaces are designed so that native clients can reuse the same transcription and refinement pipeline.
npm install
npm run devOpen http://localhost:5173 and click Push to Talk to try the browser voice capture.
Build:
npm run buildRun tests:
npm run testCAPTURE -> TRANSCRIBE -> REFINE -> PERSONALIZE -> INJECT -> SYNC
The open-source core should stay provider-neutral. Paid features should gate managed infrastructure, team controls, and convenience, not the basic ability to dictate with user-owned models.
Free:
- local transcription
- BYOK LLM cleanup
- personal dictionary
- local snippets
- no vendor lock-in
Pro:
- managed cloud STT
- faster refinement
- cross-device sync
- advanced styles
- prompt and snippet packs
Teams:
- shared dictionary
- shared snippets
- usage dashboards
- admin policy controls
- SSO-ready architecture
Enterprise:
- private deployment
- audit logs
- retention controls
- managed compliance
- custom model routing
See docs/ROADMAP.md for the full prioritised roadmap. High-level phases:
- Browser MVP (Now) — Web Speech API capture, terminal bridge, refinement pipeline
- Desktop Bridge (July 2026) — Tauri/Electron wrapper, global hotkeys, local Whisper
- Cloud & Team (August 2026) — Managed STT, LLM gateway, team sync, Stripe billing
- Enterprise & Moats (Q4 2026) — SOC 2, HIPAA, SSO, private deployment
- Scale & Exit (2027) — 5,000+ seats, $1M+ ARR, acquisition conversations
docs/REPORT_2026-06-03.md— Comprehensive strategic report with competitor audit, market analysis ($22.5B voice recognition market), profit margins (80% gross), five moats, name risk analysis, voice-training strategy, and sneaky launch tacticsAGENTS.md— System instructions for AI agents working on this codebase
AGPL-3.0-or-later. The core stays open. SaaS wrappers and managed features can be proprietary.