Frequently asked questions

Is it free?

The app is free and open source (MIT). Transcription is free and runs on your machine. The only money involved is optional: if you choose a cloud LLM, you pay that provider per use. Use Ollama locally and it's free end to end.

What data leaves my machine?

Your audio is never uploaded — transcription is fully local. The only things that ever leave your computer are: the app and model downloads, and — if you choose a cloud LLM — the transcript text sent to that provider to write the summary. Pick Ollama and nothing leaves at all. There is no Chronicle Keeper server, even with sync on (that's a separate server you host yourself, and it never sees your audio or keys).

Does it work offline?

Yes, once set up. After the one-time app and speech-model downloads, transcription works with no internet. A cloud LLM obviously needs a connection; a local Ollama model does not.

What languages does it handle?

The speech engine (Parakeet TDT v3) covers 25 European languages, German included. Summary quality in your language depends on the LLM you pick.

Do I need a powerful PC or a GPU?

No GPU required — transcription runs on plain CPU at several× realtime. A faster CPU and more RAM help, especially if you also run a local LLM. There's no Python or CUDA toolchain to install.

Do I have to use Craig Bot?

Chronicle Keeper is built around Craig's multi-track recordings — one audio file per speaker — because that gives the cleanest, speaker-attributed transcripts without diarization. That's the supported path.

Can I edit notes in the app?

Deliberately not. Chronicle Keeper generates the note and exports it; polishing happens in your editor or Obsidian vault. Keeping the app simple is a core design goal.

Do I need Obsidian?

No. Exports are plain Markdown with YAML frontmatter — they open in any text editor. The frontmatter is just a bonus for Obsidian users.

Which LLM gives the best summaries?

Generally, larger cloud models (e.g. Anthropic's Claude) write the most polished recaps. Local models are very usable and improving fast — try a 7–8B model with Ollama and step up to cloud only if you want more. The Codex improves name accuracy regardless of model.

Is it open source? Can I contribute?

Yes — MIT licensed. The Rust core lives in crates/ck-core inside a Tauri shell. PRs welcome; see CONTRIBUTING.md.