Module 6 of 6 · About 35 min
Search and databases: what to choose
Know when files are enough, when a database helps, and which one to pick.
Short answer: start without a database. Your files, the theme map and the links are enough for the first few hundred notes. A database becomes useful when you no longer find things by name, and want to find them by meaning: “that article about sleep and memory” when the note is titled “Walker, ch. 6”.
Two ways to search
- By words: finds the notes that contain the words you typed. Fast, exact, good with names and codes. Misses synonyms.
- By meaning (“semantic” search, with embeddings): each passage is turned into a list of numbers that represents its meaning, and nearby meanings are found even with different words. Needs a model to compute those numbers.
The best tools combine both, because meaning blurs proper names and words miss reformulations.
The options, from simplest to most demanding
| Option | For | Against | For whom |
|---|---|---|---|
| Files + theme map + links (this course) | Nothing to install, readable, Claude reads it directly | Slows down beyond a few hundred notes | Everyone, starting today |
| Obsidian search and Bases | Built in, visual, filters on note headers | For you only: Claude does not use it; words only | Browsing by hand |
| SQLite with full-text search | A single file, free, offline, fast | Words only; needs a small script | Step 2, with help |
| SQLite + vectors (sqlite-vec) | Words and meaning in one file | Needs an embedding model | Step 2, curious profiles |
| Chroma or LanceDB, on your computer | Meaning search that is easy to set up | Python, one more tool to maintain | Comfortable with installing things |
| Postgres + pgvector in Docker (what I use) | Words + meaning, robust, everything local | Docker; containers can stop without warning, so it needs monitoring | Advanced |
| Supabase or another hosted database | Nothing to install | Your personal notes in the cloud, a dependency, data protection questions | Avoid for personal content |
| Notion as a database | Familiar | Claude needs a connector to read it, and your notes are no longer files | Not for this method |
And Obsidian, compared with the rest?
| Tool | Why or why not |
|---|---|
| Obsidian (recommended) | Local Markdown files that Claude reads without a connector, links, graph, free |
| Logseq | Close to Obsidian, local, but built around bullet points rather than pages |
| Notion | Very good for teams, but online and readable by Claude only through a connector |
| Anytype | Local and private, but its own storage format rather than plain files |
| A folder + a text editor | Works perfectly with Claude, but no graph or backlinks |
| Capacities, Tana | Pleasant, but online and closed formats |
When to move to step 2
Three signs: Claude regularly answers “no note covers it” when you know one does; you have more than 300 notes; you search by idea rather than by name. Before that, a database is a tool to maintain with nothing to show for it.
Your weekly routine
- Drop what deserves keeping into
raw/and add it all at once. - Run a check and fix one or two points.
- Open the graph and link one orphan.
Ten minutes a week. That is all it takes for the brain to keep compounding.
To do in your vault
Tick as you go: it is kept in this browser.