The main README is the pitch — the problem, the feature list, install, the MCP tool table, the benchmarks. This folder is the depth behind it.
If you got here from a feature bullet in the README, you're in the right place. Each page below stands on its own: it opens with the problem it solves, then how remindb solves it, in plain language.
Start here if you want to understand the design rather than just use it.
- The node tree — why memory is a tree of typed nodes, not a folder of files. The
MemoryTreeorientation call. - Temperature — hot/cold decay, and the in-band cold-node summarization loop that fires without a cron.
- Versioning — snapshots, diffs, and the tiny-payload resync (
MemoryDelta/MemoryDiff/MemoryHistory). - Search — FTS5 ranked search, the OR-rewrite you must know, and budgeted fetching.
- TOON encoding — when uniform structure stores ~40% smaller, and why prose stays plain.
- MathML → LaTeX — the same ≥15% rule applied to math: MathML in HTML becomes compact LaTeX.
- Knowledge graph —
[[wiki-link]]edges, weighted traversal, and why relations are a sideband. - Architecture — the layer-by-layer map of the whole pipeline.
- CLI reference — every subcommand:
compile,serve,inspect,bench,doctor,update. - Configuration — the
.remindb/directory:config.jsonfeature blocks,ignore,temperatures.json,pinned.
Same rules as the rest of the project — see CONTRIBUTING.md. Docs changes go on a docs/<slug> branch off dev. The voice here is deliberately human: first person, problem-then-fix, no manual-speak. Match it.