The graph explorer

mait-code graph is a full-screen, read-only window onto the knowledge graph — the entities and relationships the observe hook quietly accumulates alongside your memories. Where the memory browser shows what the companion remembers, this shows what it has connected: pick an entity and see its immediate neighbourhood, either as a node-link diagram or as a flat relationship table, with every edge's free-text context a glance away.

Why it exists

Every session, the observe hook extracts entities (people, projects, tools, services, concepts, organisations) and typed relationships between them into memory.db. The text verbs — mc-tool-memory entities and relationships — could already list them, but lists hide the one thing a graph is for: shape. That your projects share infrastructure, that one tool sits under half your stack, that a person connects two clusters. The explorer makes the neighbourhood visible, and surfaces the relationship context sentences — the richest field the graph holds, which nothing else displays.

It is deliberately ego-centric: you always look at one entity's 1-hop neighbourhood, never the whole graph at once. The topology decides this — the graph is hub-and-spoke with a long noise tail, so a whole-graph view would be an unreadable hairball while any single neighbourhood renders cleanly.

The layout

• The masthead — the shared brand banner, labelled Graph, with the entity count, the current centre and its link count, and the filter state folded into the subtitle.
• The body — a filter box over the entity list on the left (ordered by mention count, each row glyph-typed), the neighbourhood in the centre (the widest share), and the detail pane on the right.
• The footer — the live key hints.

Each entity type has a glyph and a theme colour: ● person, ◆ project, ▲ tool, ■ service, ○ concept, ★ org, · unknown. The colours follow the active theme, so the graph re-skins with Ctrl+P like every other mait-code TUI.

Two views of a neighbourhood

The centre pane has two interchangeable renderings — t swaps them:

• The graph view draws the neighbourhood as a node-link diagram (via netext): boxes coloured by entity type, the centre entity inverted, arrows showing direction. Edge labels appear only on small neighbourhoods (≤15 nodes) — the renderer draws them with no collision avoidance, so at hub density they would overdraw each other. +/- zoom; the view starts auto-fitted.
• The table view lists one relationship per row: glyph-typed entities, a directional ─type─▶ column, and the context line (clipped — the detail pane has it in full). At hub density this reads better than any diagram, and it is the only view that shows every edge when two entities are connected by more than one relationship type (the diagram draws one line per pair).

Walking the graph

Navigation is the same in both views:

• Scroll the entity list and the explorer follows: the detail pane previews each entity at once, and the neighbourhood re-centres when the cursor rests for a beat (hub layouts are too heavy to redraw on every arrow press). Enter re-centres immediately.
• In the graph view, click any node to re-centre on it; click an edge to read its context in the detail pane.
• In the table view, moving the cursor previews each relationship's context; Enter on a row walks to the entity at its other end.

The detail pane always shows the selection: an entity's type, mention count, link count and seen window — or a relationship's endpoints, seen window, and its full context text.

The noise filter

Most of the graph's entities are single-mention dust — session ephemera the extractor noticed once and never again. By default the entity list hides anything with fewer than two mentions or no relationships; a toggles the full, unfiltered list when you need to find something faint. Nothing is ever deleted by the explorer — the filter is a view, not a cleanup. (For actual cleanup, mc-tool-memory entities merge folds duplicate entities together — see How memory works.)

Keys

Key	Action
/	Focus the filter
t	Swap graph ⇄ table
a	Toggle the noise filter (all ⇄ connected)
+ (or =) / -	Zoom the graph view
r	Reload from the database
1 / 2 / 3	Focus list / centre / detail
Ctrl+P	Command palette (incl. theme switching)
?	Key cheat-sheet
Esc	Back to the list; from the list, quit
q	Quit

Non-TTY fallback

Like the other TUIs, mait-code graph only opens the explorer on a TTY. Piped or in CI it prints the graph's hubs instead — the connected entities with their types, mention counts and link counts — so the command always answers something.

See also

• How memory works — how entities and relationships are extracted, and the canonical type vocabularies.
• The memory browser — the entries the graph is derived from.
• The review queue — keeping the curated tier fresh.