All use cases
Use case · Knowledge
Knowledge base
Turn your scattered docs, meetings, and notes into a searchable second brain. The vector store does the indexing; you keep editing markdown.
Knowledge base
0:24 / 1:12
Preview
How Froots handles a KB
Drop markdown into workspace/kb/ and the indexer chunks it by heading, embeds each chunk with BGE-small (384 dimensions), and stores it in a libsql DiskANN index. On every prompt, the top-K most similar chunks (default 8, similarity ≥ 0.45) are injected into the system prompt.
There’s no extra setup. Save a file, the watcher notices, the indexer runs in the background.
Skills that pair well
- `obsidian` — point Froots at your vault and the indexer covers it.
- `notion` — pull pages out of Notion into local markdown for indexing.
- `summarize` — compress long sources into shorter chunks before they hit the KB.
- `openai-whisper` — transcribe audio sources first so the text can be indexed.
Where this beats a cloud KB
- It’s local. No upload, no third party, no annual cost. The libsql file lives next to the app.
- It’s plain markdown. You can edit, diff, and version-control it.
- It’s queryable by your agent. Not just text-search — the model gets the relevant chunks in context every turn.
The KB and your conversations share the same index. A note you wrote last month and a chat from this morning are both surfaced by the same retrieval call, ranked by similarity.
What’s missing (today)
Recency decay, citations-back-to-source-line, and a polished “ask the KB” query UI aren’t shipped yet. Today this is a workflow built on chat + indexed markdown, not a dedicated product surface.