Quick Start

Five minutes from zero to a searchable, local knowledge base.

1. Install kura.

   On macOS with Homebrew:

   brew install kechol/tap/kura

   The formula pulls in the Homebrew SQLite kura needs to load its search extensions. On other platforms, grab the archive for your system from Releases — see Installation for the details.

2. Initialize the store.

   kura init      # create ~/.kura, download the tokenizer, create the DB
   kura doctor    # verify SQLite, extensions, LLM providers, DB integrity

   kura init downloads the Japanese morphological tokenizer (sqlite-vaporetto, ~6.5 MB) from GitHub Releases with SHA256 verification. On platforms it doesn’t support, kura falls back to a trigram tokenizer automatically.

3. Add a few documents.

   kura add notes/sqlite-wal.md --tags tech/db/sqlite
   echo "# Today's note" | kura add - --title "Today's note"
   kura clip https://example.com/article    # capture a web page, cleaned by the LLM, filed under clips/
   kura mv "Today's note" --path db/sqlite  # optionally file it under a path later

   A document path (--path) is always optional — documents without one live at the bucket root, which works fine as an inbox.

4. Search three ways.

   kura search "WAL checkpoint"            # keyword (FTS5 BM25, < 100ms)
   kura vsearch "how writes stay readable" # semantic (vector KNN)
   kura query "SQLite concurrency"         # hybrid + rerank (best quality)

   Keyword search always works. vsearch and query use a local LLM when one is available and degrade gracefully when it isn’t — see Search.

5. Read, link, and browse.

   kura get "Today's note"      # by doc key, #key, full path, or title
   kura link ls "Today's note"  # outlinks, backlinks, 2-hop links
   kura browser                 # http://127.0.0.1:7578 — viewer, editor, graph

Notation inside documents

• [[Title]] / [[Title|display text]] — wiki links. Write the link first and it connects automatically when the target page is created later. If two documents share a title, [[db/sqlite/Title]] (full path) pins the link to one of them.
• #tech/db/sqlite — hierarchical tags, extracted when you save.

Next steps

• Concept — why kura refuses forced filing.
• Search — how the three search modes work.
• AI Agents (MCP) — wire kura into Claude Code and other agents.
• Configuration — models, buckets, and environment variables.