Privacy & storage
What runs locally (everything by default), what touches the network, and where data lives on disk.
Network access, exhaustively
By default Bungu makes network requests in exactly two cases:
- Model downloads — a one-time fetch of the AI models, when you ask for them.
- Link checking — right-click an external link → “Check this link” performs a HEAD/GET request. Opt-in, per link.
That’s it. Everything the renderer needs (code highlighting, math, diagrams) is bundled in the app — no CDN. Chat inference, semantic search, dictation, and AI commit messages all run on-device.
If you configure a cloud provider with your own API key, chat requests (your prompt plus document/workspace context) go to that provider while it’s selected. Local is the default.
There is no telemetry and no account.
Where things live
| Path | Contents |
|---|---|
| Your files | Wherever they already are — Bungu edits markdown in place, no proprietary library or database |
<workspace>/.bungu/index/ | The local semantic index (vectors, chunk metadata). Safe to delete; it rebuilds |
<workspace>/.bungu/chats/ | Per-document chat threads, one JSON file per doc |
~/.bungu/snippets/, ~/.bungu/templates/ | Your snippets and templates, plain markdown |
~/Library/Application Support/Bungu/ | Global chat history fallback |
~/Library/Caches/models/ | Downloaded model weights |
~/Library/Logs/Bungu/ | Local trace logs (ai-trace.log, rag-trace.log) — never transmitted |
| macOS Keychain | Cloud API keys, if you add any |
.bungu/ is appended to .gitignore automatically in git repos.
Permissions
Dictation asks for microphone and speech-recognition access — both Apple frameworks, processed on-device. Bungu requests nothing else.