graphify · mapping its own core library

The whole thing, and what actually stays in memory

Real output from running graphify on graphify: tree-sitter AST, code-only, no API key. Not a mockup — the graph below, and the numbers next to it, come straight out of this run.

Left is the entire thing, force-directed and clickable. Right is what an assistant would actually load into context to answer one question about it.

Full graph

graphify-out/graph.html

2,053 nodes· 4,052 edges· 124 communities· 80 source files

EXTRACTED — explicit in source (89%) INFERRED — resolved by graphify (11%)

24.1×

fewer tokens per query, vs. reading the raw corpus

corpus, naive read~136,866 tok

avg. scoped query~5,680 tok

Reduction by question

"what connects the data layer to the api"

63.6×

"what is the main entry point"

34.5×

"how are errors handled"

23.5×

"what are the core abstractions"

12.7×

What gets loaded, per question

graphify explain

$ graphify explain "_read_text()"

Node: _read_text()

Source: extractors/base.py L65

Community: _read_text

Degree: 82

Connections (82):

<-- _collect_js_symbol_resolution_facts() [calls] INFERRED

<-- _ts_walk_class_members() [calls] INFERRED

<-- _js_extra_walk() [calls] INFERRED

<-- _require_imports_js() [calls] INFERRED

<-- _java_method_receiver_types() [calls] INFERRED

<-- _java_receiver_type_name() [calls] INFERRED

… 76 more edges — the node and its neighborhood, nothing else