Compute node embeddings using random walks + Skip-gram. - `dim`: embedding dimension (default 64) - `walks_per_node`: random walks starting from each

Incremental detection: compares against a stored manifest and returns only changed / new files. Uses content hashes to detect modifications in existin

Export the graph to GraphML format.

Export wiki-style markdown documentation. Generates a `wiki/` directory with: - `index.md` — table of contents - One `.md` file per community - One `

Extract text between two delimiters in a string.

Filter changed paths to only include relevant source files.

Generate a comprehensive markdown analysis report.

Generic platform install — just writes AGENTS.md section.

Generic platform uninstall — just removes AGENTS.md section.

Resolve a Dart import. Labels like `"import 'package:foo/bar.dart'"` or `"import 'bar.dart'"`. Extracts the file stem from the path.

Ensure a path stays within an allowed directory (no `../` traversal). Uses `canonicalize` to resolve symlinks and relative components. Returns `PathN

Persist the current manifest to disk.

Score nodes by relevance to search terms using a pre-built SearchIndex. Builds a fresh index per call. Callers doing repeated searches should use [`S

Stream the graph as NetworkX `node_link_data` JSON directly to a writer. Serialize to the NetworkX `node_link_data` JSON format, writing to the provi

Build a cache filename from a file path relative to `root`. The key is `{sha256}.json` where the hash is computed over the file content, so any chang

Find nodes that bridge multiple communities. A bridge node has a high ratio of cross-community edges to total edges. Returns up to `top_n` nodes sort

Find connected components within a subset of nodes using BFS.

Compute the SHA256 hex digest of arbitrary bytes.

Build a hashmap for fast extension lookup (cached).

Estimate the number of tokens in a string. Uses the approximation: 1 token ≈ 4 characters.

Get basic statistics about the graph.

Check whether graphify hooks are installed in the repository at `repo_root`. Returns a human-readable status string.

Is this a method stub (.method_name() or isolated fn()?

Read `.graphifyignore` from `root` and return the raw pattern strings. Returns an empty vec if the file does not exist.

Create a temporary project tree for integration tests.

Sum of weights of edges incident to a node.

Resolve a Go import to target entities. Go import labels are like `"fmt"`, `"net/http"`, or `"myproject/pkg/utils"`. Handles dot imports (`import . "

Resolve Python `import` / `from ... import` edges to actual module/function nodes discovered across files. Also handles `from x import *` by expandin

Run the benchmark suite on the graph at `graph_path`. # Arguments `graph_path` - Path to the graph JSON file. `corpus_words` - Optional word count of

Sanitize a label for safe use in HTML/GraphML output. - Strips control characters - Truncates to 256 characters - HTML-escapes `&`, `<`, `>`, `"`, an

Save a query result (question + answer) to the memory directory. Used by the `save-result` CLI command to persist LLM query results for future refere

Simulate a query against the graph and estimate context tokens needed. For each query, we find matching nodes and gather their neighborhood, then mea

Build an adjacency list from the KnowledgeGraph for efficient lookups.

Build a reverse lookup from node ID to community ID.

Remove all cached files from the given cache directory.

Simple day counter: days since 2020-01-01 from an ISO date string. Uses the same calculation as [`chrono_days_since_epoch`] to avoid offset from using

Internal detect that optionally computes content hashes during the walk to avoid double I/O when doing incremental detection. When `compute_hashes` i

DFS traversal from start nodes up to a maximum depth. Returns `(visited_nodes, edges_traversed)` where edges are `(source, target)` pairs.

Export a split HTML visualization into `output_dir/html/`. Generates: - `html/index.html` — overview page where each community is a single super-node

Find top-N most similar node pairs by cosine similarity of embeddings.

Get a mutable reference to a node by its string ID.

Check if an IP address is private, loopback, link-local, or reserved.

Returns `true` if `path` (relative to `root`) matches any ignore pattern.

Check whether the path lives inside an Xcode asset catalog.

Return the language name for a file extension (e.g. `".py"` → `"python"`).

Load configuration from `graphify-rs.toml` in the given directory. Returns default config if file doesn't exist or can't be parsed.

Load a knowledge graph from a JSON file.

Load a previously-saved manifest from disk.

Phase 1: Greedy modularity optimization (Louvain move phase). Iterates over nodes and moves each to the neighboring community that yields the greates

Normalize a file path to POSIX style: forward slashes, no leading `./`. Note: does not resolve `..` segments — assumes both tools produce consistently

Print a human-readable benchmark report.

Select the most important nodes for visualization when the graph is too large. Strategy: 1. Include top N nodes by degree (hub nodes) 2. Include at l

Phase 2: Leiden refinement. For each community, find its connected components. If a community is internally disconnected, split it — move each discon

Resolve a JS/TS import label to target entities. Import labels can be: - `"module/ExportedName"` (named import from module) - `"DefaultName"` (defaul

Resolve a Rust `use` import to target entities. Handles `pub use` re-exports, glob imports (`use foo::*`), and specific type imports (`use crate::mod

Sanitize a label for use as both a filename and a `[[wikilink]]` target. Truncates to [`graphify_core::MAX_FILENAME_BYTES`] to avoid "File name too l

Try to split an oversized community by running partition on its subgraph.

Compute total weight of edges in the graph (sum of all edge weights).

Truncate a string to at most `max_bytes` bytes while preserving UTF-8 validity.

Uninstall graphify git hooks from the repository at `repo_root`. Removes the graphify marker block from each managed hook file. If the resulting file

Architecture-level summary: group by directory, aggregate edges.

Build the inverted index from a knowledge graph. Token weights: - Label token: `2.0 + ln_1p(degree) * 0.1` - Id token: `1.0 + ln_1p(degree) * 0.1`