Exploration

The graph is not a black box that only briefs your agent. You can interrogate it directly: why is this here? what breaks if I touch that? what did we decide about X?

Beneath the decisions, patterns, and progress entries sits a combined knowledge + code graph — in this repository, 802 edges: 45 hand-made reasoning links plus 757 derived from anchors, tag overlap, and git co-change, spanning 67 code nodes. The commands below are how you walk it.

"Why is this here?" — engrams decision search

What you run: 

engrams decision search "ontology"
engrams decision search "policy engine" --snippets --limit 5
engrams decision search "workspace isolation" --all

What you get: full-text search across every decision the project has ever logged, in milliseconds. --snippets adds highlighted match context, --limit (default 10) controls how many results come back, and --all includes superseded decisions — the ones that were later replaced, with the lineage intact.

Why it's valuable: "what did we decide about X?" is the question that usually spawns a thirty-minute archaeology dig through chat history and commit messages. Here it's one query, and the answer carries its rationale and status — including whether it's still true.

"What breaks if I touch this?" — engrams graph

What you run: 

engrams graph stats                                   # the shape of the graph
engrams graph central --limit 10                      # what everything depends on (PageRank)
engrams graph clusters                                # connected components
engrams graph neighbors --node decision:44 --depth 2  # local blast radius
engrams graph chain --node decision:22 --rel depends_on   # what revisiting X cascades into
engrams graph path --from decision:44 --to decision:45    # how two things connect

What you get:

  • stats — node/edge counts, density, components, orphans, and degree stats. The one-glance answer to "how mature is this graph?"
  • central — PageRank centrality ranking (--type restricts to one node type). The nodes everything else leans on; touch them carefully.
  • clusters — the connected components, so you can see which subsystems of reasoning hang together and which are islands.
  • neighbors — every node within --depth hops of a node, optionally restricted to one --rel type. The immediate blast radius of a change.
  • chain — transitive closure over a canonical transitive relationship (supersedes, depends_on, part_of, refines). Literally the "what breaks if I revisit X" query: follow the chain and see everything downstream.
  • path — the shortest path between any two nodes. When two pieces of the project seem unrelated, this shows the reasoning that connects them.

Why it's valuable: code tells you what depends on a module. The graph tells you what depends on a decision — which patterns were built on it, which later decisions refined it, which work assumed it. Before you reverse a three-week-old call, graph chain shows you exactly what you're implicitly reversing with it.

"Give me the record" — engrams export

What you run: 

engrams export                       # dumps to ./engrams_export
engrams export --path ./docs/adr     # somewhere reviewable

What you get: the entire database serialized to clean, structured Markdown files — decisions, patterns, progress, context — ready to commit to git.

Why it's valuable: the record becomes reviewable, diffable, and collaborative. Architectural decisions show up as Markdown in PRs; teammates pull and engrams import to sync their local database. The graph stops being a file on your machine and becomes part of the repository's history.

Derived edges do the legwork Most of the graph isn't hand-linked. engrams graph rebuild recomputes derived edges from anchors, tag overlap, and git co-change, and engrams graph ingest incrementally adds co-change edges as you commit — so exploration gets richer the longer the project runs, without anyone maintaining links by hand.

Keep exploring

  • For You — the human's relationship with the advisor
  • Briefing — start every session oriented
  • Enforcement — standards enforced before violations land
  • The Advisor — how the graph you're exploring came to be