Faden

Desktop application for qualitative interview research. Audio transcription via Gemini, manual and AI-assisted tagging with a three-level codebook (cluster -> category -> tag), per-second audio-to-transcript linking, and five export formats (CSV, Markdown, REFI-QDA, stats, codebook).

Status: Alpha. Built for personal research use. May be open-sourced once it stabilises.

Linux / Wayland note: Sorry — Tauri currently forces client-side window decorations in some Wayland setups, so Faden may show an ugly custom titlebar instead of your compositor defaults. Upstream bug: tauri-apps/tao#1046

Website: https://faden.space

Features

• Project-per-folder. Each project is a directory with project.sqlite, a media/ folder for original audio, and a cache/ folder for transcoded WAV chunks.
• Multi-format import: audio only, transcript text, transcript JSON, or audio plus matching transcript.
• Cloud transcription via Google's Gemini models with chunked retry and a progress bar per chunk.
• Three-level codebook (cluster / category / tag) with colors, descriptions, and reorder.
• Selection-based tagging with per-quote memos and shared-tag rollups.
• Audio playback driven by transcript clicks, plus loop-current-span.
• AI assists, all explicitly user-triggered (no background calls): codebook generation, pre-tagging, "find more like this".
• Export to CSV (per-quote), Markdown, REFI-QDA XML, stats reports (frequencies and co-occurrences), and standalone codebook (JSON or CSV).
• English and Czech UI.

Install (from source)

Prerequisites

• Rust toolchain (rustup)
• Node.js 18+
• ffmpeg and ffprobe installed on the system
• A Gemini API key (https://aistudio.google.com/apikey)
• System WebKitGTK on Linux (Tauri runtime)

Setup

git clone <repo-url> faden
cd faden
npm install
npm run tauri dev           # development build with hot reload

To produce a release bundle:

npm run tauri build

Output bundles land in src-tauri/target/release/bundle/ (deb, AppImage, or platform-specific installers).

First-run walkthrough

1. Launch the app. The project picker opens. Click New project, choose an empty folder. The picker also lists recent projects with inline rename / remove.
2. Add your API key. Open Settings (gear icon in the workspace). Paste your Gemini key. Faden stores it in your OS keychain / credential store, not in plaintext project settings. Optionally set the default transcription / AI model and UI language.
3. Import an interview. In the left pane click + From audio, pick an audio file. The app transcodes a working copy via the system ffmpeg and creates a new interview row.
4. Transcribe. Click Transcribe next to the interview. Long files are chunked; progress is shown per chunk. Failed chunks can be retried.
5. Build a codebook. In the left pane add clusters, then categories, then tags. Or use AI -> Generate codebook to draft one from selected interviews.
6. Tag quotes. Select transcript text and apply a tag. Optionally add a memo. Tagged spans appear in the right pane with their tags and the underlying audio loop.
7. AI pre-tag (optional). Pick an interview, open the AI panel, choose Pre-tag. Review the staged proposals and accept individually or in bulk.
8. Export. From the workspace menu choose Export. Pick scope (current interview / whole project / by tag), formats (CSV, Markdown, REFI-QDA, stats, codebook), and an output directory.

Settings overview

• Gemini API key - required for any AI feature. Stored in the OS keychain / credential store.
• Default transcription model / Default AI tagging model - the model identifiers to pass to Gemini. Default is gemini-3-flash-preview.
• UI language - Auto (OS default), English, or Czech.
• Prompts - per-project overrides for transcription, codebook generation, pre-tagging, and find-more system / user prompts. Empty fields fall back to the built-in defaults.

Project layout on disk

my-project/
  project.sqlite      # codebook, interviews, segments, spans, settings
  media/              # imported originals (audio, transcripts)
  cache/              # transcoded WAV chunks, transient files

Projects are portable: zip the folder, move it to another machine with the same app installed, and reopen via Open folder....

Development

Prerequisites

• Rust toolchain (rustup)
• Node.js 18+
• ffmpeg + ffprobe
• WebKitGTK + libsoup3 (Linux desktops)

Common commands

npm install                     # install JS deps
npm run dev                     # vite-only dev server
npm run tauri dev               # full Tauri dev shell
npm run build                   # tsc + vite production frontend build
npm run tauri build             # signed/unsigned platform installer
cargo test --manifest-path src-tauri/Cargo.toml

Linux packaging

Linux builds use the system ffmpeg / ffprobe and Arch packaging is handled via the AUR metadata generated in scripts/generate-aur-package.sh and scripts/generate-aur-git-package.sh.

Testing

• Rust backend: cargo test from src-tauri/.
• Frontend type-check: tsc --noEmit.
• No end-to-end harness yet; manual smoke list lives in docs/superpowers/plans/.

Architecture pointers

• src-tauri/ - Rust backend, Tauri commands, SQLite layer (db/queries/), AI providers (ai/), transcription pipeline (transcription/), export adapters (export/).
• src/ - React 19 frontend. State in Jotai (src/state/), routing via TanStack Router (src/router.ts). No CSS-in-JS, no Tailwind - plain CSS modules.
• src/ipc/ - typed wrappers around invoke(...) Tauri commands. Always add a wrapper rather than calling invoke from view code.
• src/i18n/{en,cs}.json - translations. Files must stay in structural parity; an audit script lives in the polish plan.

Roadmap

See docs/superpowers/plans/ for per-feature plans (01 through 08) and docs/superpowers/roadmap.md (if present) for the high-level outline.

License

MIT - see LICENSE.