FlowBridge (Wispr Flow Clone)

Wanted a dictation tool that doesn't just transcribe — it cleans up after itself, in two languages.

Wispr Flow is a commercial speech-to-text app. I wanted the same core idea — hold a hotkey, talk instead of type, get clean text wherever the cursor is — but built and owned by me, with support for how I actually work: switching between English and Urdu. So I built my own version, FlowBridge, as a personal project.

Directed the entire build with Codex, then wired in real infrastructure around it.

• Built end-to-end with Codex — rather than hand-writing the app, I directed Codex (an AI coding agent) to build it, iterating until the workflow felt right: hold Ctrl+Space, speak, release, and the cleaned-up text lands wherever I was typing
• Groq primary, Gemini fallback — speech-to-text runs on Groq for speed, with Gemini wired in as an automatic fallback so dictation doesn't just stop if one service has issues
• Bilingual English + Urdu support — the app transcribes and cleans up both languages, including mixed sessions, rather than being English-only like most dictation tools
• Built-in usage analytics — an Insights page tracks words dictated, sessions, speaking speed (WPM), average latency, language mix, and which apps I dictate into most (Chrome, Claude, ChatGPT)
• Custom dictionary — a dictionary section for words/terms the model should always get right, rather than fighting the same misheard word every time
• Privacy-first architecture — audio is never saved to disk; local history and API keys are encrypted for my Windows account; transcript retention and local database size are both configurable, with one-click options to clear temporary data or open the data folder