Struct KuzuGraphStore 

pub struct KuzuGraphStore { /* private fields */ }

Local KuzuDB-backed implementation of GraphStore.

One database file per repo (graph.kuzu), with per-branch node/edge tables inside it. A fresh Connection is created for each operation so we avoid the self-referential lifetime that Mutex<Connection<'db>> would require.

Implementations

impl KuzuGraphStore

pub fn open(repo_root: &Path) -> Result<Self>

Open (or create) the graph database for the repo at repo_root.

Existing data with a different schema is never deleted as a side effect of a query, hook, or server startup. An explicit gcx init performs the rebuild through Self::open_for_init.

pub fn open_for_daemon(repo_root: &Path) -> Result<Self>

Open the graph as the shared local MCP owner.

pub fn open_for_init(repo_root: &Path) -> Result<Self>

Open the graph for an explicit initialization operation, rebuilding an incompatible store only while exclusive repository ownership is held.

Trait Implementations

impl GraphStore for KuzuGraphStore

fn apply_diff(&mut self, branch: &str, diff: &GraphDiff) -> Result<()>

Apply an incremental diff to the named branch’s graph.

fn lookup_symbol( &self, branch: &str, name: &str, fuzzy: bool, ) -> Result<Vec<Node>>

Find all nodes matching name on branch. When fuzzy is true, matches any node whose name contains name (case-sensitive substring). When false, exact match only.

fn find_callers(&self, branch: &str, function_name: &str) -> Result<Vec<Node>>

Find all call-site nodes whose outgoing Calls edge points to a node named function_name on branch (single hop).

fn find_callers_with_confidence( &self, branch: &str, function_name: &str, ) -> Result<Vec<(Node, EdgeConfidence)>>

Like find_callers but also returns each caller’s edge confidence. The default conservatively marks all callers as Inferred; backends should override with a query that reads e.confidence from the edge.

fn find_callers_by_id_with_confidence( &self, branch: &str, target_id: &str, ) -> Result<Vec<(Node, EdgeConfidence)>>

Find direct callers for one exact target node ID, including edge confidence. Agent-facing queries use this after symbol disambiguation so common names such as get cannot merge unrelated call graphs.

fn find_callers_deep( &self, branch: &str, function_name: &str, depth: u8, ) -> Result<CallersDeep>

Multi-hop BFS: find callers up to depth hops away. Returns callers grouped by hop distance (1..=depth). depth is capped at 5 to prevent runaway queries.

fn symbol_context(&self, branch: &str, name: &str) -> Result<SymbolContext>

Return a 360° view of a symbol: its definition, direct callers, direct callees, and nodes that reference it via Uses edges.

fn list_definitions(&self, branch: &str, file: &Path) -> Result<Vec<Node>>

List all top-level definitions in file on branch.

fn branch_diff(&self, from: &str, to: &str) -> Result<GraphDiff>

Return the graph delta between two branches as a GraphDiff. Nodes/edges present in to but not from are in added_*. Nodes/edges present in from but not to are in removed_*.

fn list_all_nodes(&self, branch: &str) -> Result<Vec<Node>>

Return all nodes in branch’s graph.

fn search_nodes( &self, branch: &str, query: &str, limit: usize, ) -> Result<Vec<Node>>

Return nodes whose name or qualified_name contains query (case- sensitive substring), up to limit results. Implementations should push the filter to the store rather than scanning all nodes in memory.

fn get_nodes_by_ids(&self, branch: &str, ids: &[String]) -> Result<Vec<Node>>

Resolve a set of node IDs to full nodes. Order is not guaranteed; IDs that don’t exist on branch are silently skipped.

fn list_all_edges(&self, branch: &str) -> Result<Vec<Edge>>

Return all edges in branch’s graph.

fn list_nodes_page( &self, branch: &str, offset: usize, limit: usize, ) -> Result<Vec<Node>>

Return a deterministic page of nodes ordered by stable node ID.

fn list_edges_page( &self, branch: &str, offset: usize, limit: usize, ) -> Result<Vec<Edge>>

Return a deterministic page of edges ordered by source, destination, kind, and source line. See GraphStore::list_nodes_page.

fn search_by_attributes( &self, branch: &str, filter: &AttributeFilter, limit: usize, ) -> Result<Vec<Node>>

Find nodes matching a structural filter (kind, async, visibility, complexity range, name substring), up to limit results.

fn graph_stats(&self, branch: &str) -> Result<GraphStats>

Aggregate node/edge counts (total + per-kind) for branch.

fn find_callees( &self, branch: &str, function_name: &str, depth: u8, ) -> Result<CallersDeep>

Find all functions/methods called by function_name up to depth hops. Returns callees grouped by hop distance (1..=depth). Capped at 5.

fn find_implementors( &self, branch: &str, trait_or_interface_name: &str, ) -> Result<Vec<Node>>

Find all structs/classes that implement/inherit trait_or_interface_name.

fn find_type_usages(&self, branch: &str, type_name: &str) -> Result<Vec<Node>>

Find functions/methods that reference a type named type_name as a parameter or return type (following Uses edges). Answers “where is type T used in a signature” — the type-level analogue of find_callers.

fn find_call_sites( &self, branch: &str, function_name: &str, ) -> Result<Vec<CallSite>>

Find every call site of the function named function_name: the calling symbol plus the source line of each call expression (following Calls edges). Where find_callers returns only the calling symbols, this also pinpoints the line each call happens on.

fn find_importers(&self, branch: &str, symbol_name: &str) -> Result<Vec<Node>>

Find the module/file nodes that import a symbol named symbol_name (following Imports edges). Answers “who imports X”.

fn type_hierarchy(&self, branch: &str, name: &str) -> Result<TypeHierarchy>

Return both directions of the type relation for name: the types it implements/extends (supertypes) and the types that implement/extend it (subtypes), following Implements and Inherits edges.

fn trace_path(&self, branch: &str, from: &str, to: &str) -> Result<Vec<Node>>

Find all call paths between from and to using BFS. Returns at most one path (the shortest), as a sequence of nodes.

fn list_symbols_in_range( &self, branch: &str, file: &Path, start_line: u32, end_line: u32, ) -> Result<Vec<Node>>

Find all nodes in file whose span overlaps [start_line, end_line].

fn find_unused_symbols( &self, branch: &str, kind: Option<NodeKind>, ) -> Result<Vec<Node>>

Find symbols with no incoming Calls or Uses edges (potential dead code). If kind is provided, filters to only that NodeKind.

fn get_subgraph( &self, branch: &str, seed_name: &str, depth: u8, direction: &str, ) -> Result<SubGraph>

Return a subgraph centred on seed_name up to depth hops. direction: “in” (callers), “out” (callees), or “both”.

fn get_subgraph_by_id( &self, branch: &str, seed_id: &str, depth: u8, direction: &str, ) -> Result<SubGraph>

Return a subgraph around one exact seed node ID. Agent-facing queries use this after disambiguation so repeated short names cannot merge unrelated neighborhoods.

fn get_neighborhood_by_id( &self, branch: &str, seed_id: &str, direction: &str, limit: usize, ) -> Result<SubGraph>

Return one bounded hop around an exact seed ID. Visualization clients use repeated calls to expand large graphs without materializing an unbounded multi-hop neighborhood.

fn last_indexed_sha(&self, branch_name: &str) -> Result<Option<String>>

Last commit SHA successfully indexed for branch. None if the branch has never been indexed.

fn set_last_indexed_sha(&mut self, branch_name: &str, sha: &str) -> Result<()>

Persist the commit SHA after a successful index run.

fn list_edges_by_kind( &self, branch: &str, kind: EdgeKind, ) -> Result<Vec<Edge>, GitCortexError>

Return edges of a specific kind in branch’s graph. The default filters list_all_edges in-memory; backends should override with a WHERE-clause push-down for large graphs.

fn module_dependencies( &self, branch: &str, module_name: &str, ) -> Result<Vec<Node>, GitCortexError>

Return the in-repo modules that a module named module_name depends on, resolved by following its Imports edges to the defining module of each imported symbol. Answers “what does this module depend on”.