sqlitegraph 0.4.0

Embedded graph database with HNSW vector search — Python bindings to the sqlitegraph Rust crate.

sqlitegraph

Python bindings to the sqlitegraph embedded graph database. Storage, graph algorithms, and HNSW vector search run in a reviewed Rust core; this package is the Pythonic surface.

Alpha — API subject to change before 1.0.

Install

pip install sqlitegraph

Quick start

from sqlitegraph import Graph

g = Graph.open_in_memory()
alice = g.add_node(kind="User", name="Alice", data={"age": 30})
order = g.add_node(kind="Order", name="Order-123")
g.add_edge(alice, order, "placed")

print(g.neighbors(alice))

Query language

Graph.query() exposes the same Cypher-inspired language as the CLI:

from sqlitegraph import Graph

g = Graph.open_in_memory()
alice = g.add_node(kind="User", name="Alice", data={"age": 30})
bob = g.add_node(kind="User", name="Bob", data={"age": 31})
g.add_edge(alice, bob, "KNOWS")

result = g.query("MATCH (a:User)-[:KNOWS]->(b:User) RETURN a.name, b.name")
print(result["results"])

Supported query features include node scans, edge traversal, multi-hop chains, star/multi-pattern joins, variable-depth edges, WHERE with regex/numeric operators and parentheses, LIMIT, CREATE, SET, DELETE, and HNSW vector search through CALL db.index.vector.queryNodes(...).

Algorithms

from sqlitegraph import Graph

g = Graph.open_in_memory()
a = g.add_node(kind="Page", name="A")
b = g.add_node(kind="Page", name="B")
c = g.add_node(kind="Page", name="C")
g.add_edge(a, b, "LINKS")
g.add_edge(b, c, "LINKS")
g.add_edge(c, a, "LINKS")

print(g.pagerank(iterations=20))
print(g.connected_components())
print(g.strongly_connected_components())
print(g.label_propagation(50))
print(g.find_cycles(10))
print(g.dominators(a))

critical_path() is also available for directed acyclic graphs and returns {"path": [...], "distance": float, "path_length": int}.

HNSW vector search

from sqlitegraph import Graph

g = Graph.open_in_memory()
idx = g.create_hnsw_index("embeddings", dimension=3, metric="cosine")

idx.insert_vector([1.0, 0.8, 0.1], {"label": "graph databases"})
idx.insert_vector([0.1, 0.2, 1.0], {"label": "baking"})

print(idx.search([1.0, 0.9, 0.0], 1))
print(g.list_hnsw_indexes())

g.delete_hnsw_index("embeddings")

Examples

The examples/ directory contains runnable scripts:

Example	What it shows
01_basic_crud.py	Nodes, edges, update, delete, query by kind/pattern, degrees
02_graph_algorithms.py	BFS, k-hop, shortest path, PageRank, Louvain & label-propagation communities, connected components (WCC), strongly-connected components (SCC), cycle search, dominator tree, critical path
03_vector_search.py	HNSW index creation, insert, search, bulk insert, index listing
04_social_network.py	Realistic network: influencers (PageRank), communities, connection paths, mutual follows
05_file_backed.py	Persistent Graph.open(path), checkpoint, reopen, cleanup
06_hybrid_sqlite_hnsw_query.py	sqlite3 application rows + sqlitegraph metadata + HNSW + Graph.query() expansion

Run any example from the repo root:

cd sqlitegraph-py
source .venv/bin/activate
python examples/01_basic_crud.py