Module graph 

Knowledge graph with SQLite persistence, FTS5 search, and smart recall.

Provides a complete knowledge graph layer on top of SQLite:

• Types: GraphNode, GraphEdge, ScoredNode, GraphStats
• Schema: init_graph_schema — creates tables, FTS5, indexes
• CRUD: node/edge insert, read, update, delete
• Search: FTS5 full-text search and dynamic filtering
• Recall: smart_recall — composite scoring with recency, importance, access, FTS, graph boost
• Traversal: graph_neighbors (1-hop), related_nodes (BFS via recursive CTE)
• Algorithms: pure-Rust CSR algorithms in algo — pagerank(), connected_components(), label_propagation(), dijkstra(), jaccard_similarity()
• Lifecycle: decay_importance, tag_stale_nodes, compute_stats

All functions take &rusqlite::Connection — no hardcoded paths.

use rusqlite::Connection;
use llm_kernel::graph::{init_graph_schema, upsert_node, smart_recall, GraphNode};

let conn = Connection::open_in_memory().unwrap();
init_graph_schema(&conn).unwrap();

upsert_node(&conn, &GraphNode {
    id: "rust-ownership".into(),
    node_type: "concept".into(),
    title: "Rust Ownership Model".into(),
    body: "Ownership, borrowing, and lifetimes...".into(),
    tags: vec!["rust".into(), "memory-safety".into()],
    projects: vec!["my-project".into()],
    agents: vec![],
    created: "2026-01-01T00:00:00Z".into(),
    updated: "2026-01-01T00:00:00Z".into(),
    importance: 0.8,
    access_count: 0,
    accessed_at: String::new(),
}).unwrap();

let results = smart_recall(&conn, Some("my-project"), Some("ownership"), 5).unwrap();
for scored in &results {
    println!("{:.2} — {}", scored.score, scored.node.title);
}

Re-exports

pub use async_pool::AsyncPoolGraph;

pub use algo::CsrGraph;

pub use algo::LABEL_PROPAGATION_ITERS;

pub use algo::PAGERANK_DAMPING;

pub use algo::PAGERANK_EPS;

pub use algo::PAGERANK_ITERS;

pub use algo::SHORTEST_PATH_W_MIN;

pub use algo::adamic_adar;

pub use algo::common_neighbors;

pub use algo::connected_components;

pub use algo::dijkstra;

pub use algo::jaccard_similarity;

pub use algo::label_propagation;

pub use algo::label_propagation_default;

pub use algo::link_prediction;

pub use algo::pagerank;

pub use algo::pagerank;

pub use algo::pagerank_default;

pub use algo::pagerank_scores;

pub use algo::shortest_path;

pub use algo::shortest_path_ids;

pub use backend::GraphBackend;

pub use backend::SqliteGraph;

pub use dedup::find_duplicate;

pub use dedup::upsert_node_dedup;

pub use lifecycle::compute_stats;

pub use lifecycle::decay_importance;

pub use lifecycle::tag_stale_nodes;

pub use lifecycle::touch_node;

pub use lifecycle::touch_nodes;

pub use recall::smart_recall;

pub use schema::GRAPH_SCHEMA_VERSION;

pub use schema::init_graph_schema;

pub use schema::migrate_graph;

pub use schema::schema_version;

pub use search::query_nodes;

pub use search::search_nodes;

pub use store::append_edge;

pub use store::delete_edge;

pub use store::delete_node;

pub use store::read_edges;

pub use store::read_node;

pub use store::read_nodes;

pub use store::upsert_node;

pub use traversal::build_graph;

pub use traversal::graph_neighbors;

pub use traversal::related_nodes;

pub use types::Graph;

pub use types::GraphEdge;

pub use types::GraphNode;

pub use types::GraphNodeSummary;

pub use types::GraphStats;

pub use types::ScoredNode;

pub use types::validate_uuid;

pub use cjk::search_nodes_cjk;

pub use cjk::segment_cjk;

pub use pg::PgGraph;

Modules

algo

Graph algorithms operating on an in-memory CSR snapshot.

async_graph

Async wrappers for the knowledge graph.

async_pool

Multi-connection async pool for the knowledge graph.

backend

Backend-agnostic graph trait and SQLite implementation.

cjk

CJK-aware graph search (Rust-side segmentation; no schema change). CJK-aware graph search (Rust-side segmentation, no schema change).

dedup

Node deduplication: prevent duplicate nodes by title within a time window.

lifecycle

Node lifecycle: access tracking, importance decay, stale tagging, and stats.

pg

PostgreSQL GraphBackend (feature graph-pg). PostgreSQL GraphBackend (graph-pg feature).

recall

Smart recall with composite scoring and graph boost.

schema

Schema initialization for the knowledge graph SQLite database.

search

FTS5 full-text search for knowledge graph nodes.

store

Node and edge CRUD operations.

traversal

Graph traversal: 1-hop neighbors and BFS via recursive CTEs.

types

Core types for the knowledge graph.