Architecture — OpenFlow Help

Overview

OpenFlow is a local-first desktop voice dictation app built with Tauri v2, React/TypeScript frontend, and Rust backend with SQLite storage.

┌─────────────────────────────────┐
│          React Frontend          │
│  (TypeScript, Tailwind, lucide)  │
├─────────────────────────────────┤
│        Tauri Command API        │
├─────────────────────────────────┤
│          Rust Backend           │
│  ┌──────┐ ┌──────┐ ┌────────┐  │
│  │Audio │ │Hotkey│ │Cleanup │  │
│  │      │ │      │ │Engine  │  │
│  └──────┘ └──────┘ └────────┘  │
│  ┌──────────┐ ┌─────────────┐  │
│  │Transcr.  │ │  Clipboard   │  │
│  │Provider  │ │  & Paste     │  │
│  └──────────┘ └─────────────┘  │
│  ┌──────────────────────────┐  │
│  │      SQLite Database     │  │
│  └──────────────────────────┘  │
└─────────────────────────────────┘

Frontend

React + TypeScript with Vite
Tailwind CSS for styling (preserving Wispr Flow-inspired design)
lucide-react for icons
Custom hooks (useSettings, useSnippets, useDictionary, useHistory, useTransforms, useScratchpad) wrap Tauri invoke calls
API wrapper module (src/lib/api.ts) provides typed functions for all backend commands
Shared types in src/types/app.ts

Backend (Rust)

Modules

Module	Purpose
`db`	SQLite database initialization & migrations
`settings`	App settings CRUD (key-value store)
`snippets`	Snippet CRUD and expansion logic
`dictionary`	Dictionary term CRUD
`history`	Dictation history CRUD
`transforms`	Transform CRUD
`scratchpad`	Scratch note CRUD
`cleanup`	Text cleanup pipeline
`models`	Model status, audio devices, clipboard
`errors`	Typed error enums

Cleanup Pipeline

Raw Transcript
     │
     ├── Call Gemini API (REMOVED)
     │
     ▼
Clean Transcript
     │
     ├── RuleBasedProvider (always available)
     │   ├── Remove filler words
     │   ├── Handle corrections (I mean, actually)
     │   ├── Expand spoken punctuation
     │   ├── Apply formatting commands
     │   └── Capitalize and punctuate
     │
     └── LocalLlamaCppServerProvider (optional)
         ├── POST to http://127.0.0.1:8080/v1/chat/completions
         ├── System prompt for speech cleanup
         ├── 8 second timeout
         └── Falls back to rule-based on failure

Transcription Pipeline

Hotkey Pressed → Start Recording
                     │
Hotkey Released → Stop Recording
                     │
              Transcribe (Whisper)
                     │
              Clean Transcript
                     │
              Expand Snippets
                     │
              Save History
                     │
              Paste or Preview
                     │
              Return to Idle

App States

idle — Waiting for hotkey
listening — Recording audio
processing_transcription — Running Whisper
processing_cleanup — Running cleanup model
preview — Showing preview before paste
pasting — Inserting text into active app
success — Dictation complete
error — Something went wrong

Storage

All data is stored in local SQLite at:

Windows: %APPDATA%\openflow\openflow.db

Tables

settings — Key-value app configuration
snippets — Snippet shortcuts and expansions
dictionary_terms — Personal dictionary words
history — Dictation history entries
transforms — Transform templates
scratch_notes — Scratchpad notes

Security

No outbound network connections except localhost:8080 (for optional llama.cpp server)
No telemetry or analytics
Audio processed and discarded locally
All data encrypted at rest in SQLite