The typed knowledge graph

Not a visualization — a data structure the system uses. Recall walks it, provenance queries it, sleep clusters it.

Nodes

Edges

kind how it forms notes
citation you write a link (see below) typed: updates / contradicts / cites / relates
supersedes belief supersession (Beliefs, staleness & decay) weight 1.5, the strongest signal
semantic nearest-neighbour similarity automatic, the background tissue
entity shared entities (optional, Graphify) rebuild_entity_graph, local claude CLI, $0

How citations are written

All of these create edges — use whichever reads naturally:

<a class="wl" href="file-slug.html">file-slug</a>            wikilink, Obsidian-compatible
<a class="wl" href="file-slug.html">shown text</a> wikilink with alias
[[#42]]                  memory by id
[[D7]]                   decision by id
"see memory #42"         plain prose (English or Italian)
"as per decision #7"     plain prose

The relation type is inferred from surrounding words: "updates/supersedes…" → updates; "contradicts/smentisce…" → contradicts; "see/cfr…" → cites; otherwise relates.

Link generously. Every link is an edge; every edge makes associative recall, why, and sleep smarter. The graph is only as good as your linking habit — and the server instructs agents to build it as they store.

What runs on the graph

The local visualizer

python -m wadachi.web        # http://localhost:8420

The full brain as an interactive D3 graph: project clusters, decisions as diamonds, supersedes edges in vermilion, an "as of" time slider (temporal dimension), and click-through provenance on every decision. Local only — never exposed.