Transcribe Kit

Cross-platform desktop transcription app built with Tauri + Leptos. Transcribe audio files or record live — using local Whisper models or any OpenAI-compatible API — then post-process results with AI prompt templates. One Rust codebase, all desktop platforms.

Recording	Settings	Post-processing

Features

• Local & API transcription — run Whisper locally (tiny, base, small, large-v3-turbo) or hit any OpenAI-compatible endpoint. Automatic model download and caching for local mode.
• File import — drag in WAV, MP3, FLAC, OGG, or M4A files. Large files are auto-compressed to MP3 before API upload.
• Live recording — push-to-talk or toggle mode with a configurable global hotkey. Works even when the app is in the background.
• Meeting capture — dual-stream mode records microphone and system audio simultaneously, then mixes them into one file for transcription.
• Real-time streaming — transcription segments appear as they're produced, with progress updates.
• Standalone Process tab — apply AI prompt templates to any text (transcripts, notes, or pasted content) — no transcription required first. Ships with built-in templates (cleanup, meeting notes, summary), supports custom user templates with {{transcript}} and {{noteN}} placeholders, and lets you reference saved notes as additional context. After the first run, keep iterating with chat-style follow-ups (⌘/Ctrl+Enter to send) and save either the last reply or the full thread as a note. Run locally with a bundled llama-server sidecar (no API key needed) or via any OpenAI-compatible API.
• Notes — persistent local notes. Transcripts auto-save into Notes when transcription completes; processed results save on demand from the Process tab (last reply, or the entire thread). Notes can be loaded back into Process as input or referenced from templates as {{noteN}} slots.
• Secure API key storage — credentials are stored in the system keyring, not in config files.
• Persistent settings — provider, model, device, hotkey, and API config auto-save with debouncing.
• Cross-platform — macOS, Windows, and Linux from one codebase.

Tech Stack

Layer	Technology
Frontend	Leptos (Rust → WASM), bundled with Trunk
Desktop runtime	Tauri 2
Local transcription	whisper-rs (whisper.cpp bindings)
Local LLM	llama-server sidecar (llama.cpp, OpenAI-compatible HTTP API)
Audio I/O	CPAL for capture, Symphonia for decoding
API transport	reqwest with multipart streaming
Settings	JSON config + system keyring for secrets

Project Structure

frontend/src/         Leptos UI components and state
src-tauri/src/
  commands.rs         Tauri IPC command handlers
  transcription.rs    Transcription orchestration
  llm_engine.rs       llama-server sidecar lifecycle and chat completion
  live_recording/     Audio capture and mixing
  providers/          Backend adapters (Whisper, OpenAI, local LLM)
  settings.rs         Persistent config management
  templates.rs        Post-processing template storage
  notes.rs            Notes storage (one JSON file per note)

Local Development

Prerequisites:

• Rust stable
• wasm32-unknown-unknown target
• trunk
• tauri-cli
• cargo-make
• cmake (required to build whisper.cpp from source)
• A C/C++ compiler (Xcode Command Line Tools on macOS, MSVC on Windows, gcc/g++ on Linux)
• Platform dependencies required by Tauri

Commands:

cargo install cargo-make --locked
cargo make setup
./scripts/download-llama-server.sh   # download llama-server sidecar binary
cargo make dev

Production build:

cargo make build

Useful task shortcuts:

• cargo make setup: install the WASM target plus required CLI tools
• cargo make dev: run the Tauri desktop app in development mode
• cargo make build: build production desktop bundles
• cargo make build-frontend: build the Leptos frontend only

Why Leptos Here

This repo is intentionally Rust-first to reduce future JavaScript ecosystem maintenance. The UI is simple enough that Leptos and Trunk are a good fit, while native-heavy behavior like audio capture, hotkeys, settings persistence, and local model orchestration still live in Tauri.