post-graph: PostgreSQL-Backed Graph Database Library

A high-performance Python library for using PostgreSQL as a native graph database. It supports multi-tenant realms, pgvector similarity search across main & data history tables, automatic shadow audit logging, append-only history tables, and high-speed recursive graph traversals (CTEs).

🌟 Key Features

• Table-Per-Vertex & Table-Per-Edge Architecture: Maps graph elements directly to relational tables, taking advantage of PostgreSQL's foreign keys, indexes, and constraints.
• pgvector Similarity Search (vector_search):
  • Native HNSW vector indexing (vector(dim)) for vertex embeddings.
  • Cosine distance (<=>), Euclidean L2 (<->), and Inner Product (<#>).
  • Multi-scope searching across main tables, associated data history tables, or both combined (search_scope="both").
• Dual Isolation Multi-Tenancy:
  • Single-Schema Multi-Tenancy: Logical isolation using a realm column partition.
  • Schema-Per-Realm: Physical isolation creating dedicated PostgreSQL schema namespaces per tenant (CREATE SCHEMA IF NOT EXISTS "realm_name").
• Autogenerated BIGSERIAL Primary Keys & Computed fqid:
  • Vertex FQID: {realm}/{table_name}/{id}
  • Edge FQID: {realm}/{from_table}-{to_table}/{id} (using hyphen separator)
  • Automatically populated at PostgreSQL level via GENERATED ALWAYS AS ... STORED NOT NULL.
• Append-Only History Tables ({table_name}_data):
  • Automatically created alongside vertex and edge tables.
  • Stores timestamped JSONB payload updates (data_id, realm, id, payload, timestamp, embedding).
  • Supports vector embeddings on data records for historical semantic matching.
  • Cascades deletion when main vertex/edge is deleted (ON DELETE CASCADE).
• Shadow Audit Logging ({table_name}_audit):
  • Operates via PostgreSQL triggers to capture all INSERT, UPDATE, and DELETE events.
  • Logs old/new row state and the initiating user_id passed via session parameters (app.current_user_id).
• Advanced Graph Traversals & Cycle Detection:
  • Recursive CTE queries for neighbor exploration, path discovery, and cycle-free shortest path calculation.
  • Optional check_cycle=True raising CyclicReferenceError during edge creation.
  • Direct object-oriented traversal APIs (vertex.to(), vertex.from_(), step.vertex(), step.add_edge_to()).
• Multiple Async Client Drivers:
  • High-speed asyncpg client (AsyncPostGraph).
  • SQLAlchemy v2.0 async client (SQLAlchemyPostGraph).

📦 Installation

Install post-graph from PyPI:

# Basic installation (includes asyncpg)
pip install post-graph

# Installation with SQLAlchemy support
pip install "post-graph[sqlalchemy]"

# Installation with all optional dependencies
pip install "post-graph[all]"

Using uv:

uv add post-graph
# or with SQLAlchemy support:
uv add "post-graph[sqlalchemy]"

PostgreSQL Setup

To enable pgvector similarity search, ensure the pgvector extension is enabled in PostgreSQL:

CREATE EXTENSION IF NOT EXISTS vector;

🏗️ Database Architecture

erDiagram
    users {
        text realm PK
        bigserial id PK
        text fqid
        jsonb payload
        vector embedding
        timestamptz created_at
        timestamptz updated_at
    }
    posts {
        text realm PK
        bigserial id PK
        text fqid
        jsonb payload
        vector embedding
        timestamptz created_at
        timestamptz updated_at
    }
    likes {
        text realm PK, FK
        bigserial id PK
        text fqid
        bigint from_id FK
        bigint to_id FK
        text relation_type
        jsonb payload
        timestamptz created_at
        timestamptz updated_at
    }
    users_audit {
        bigint audit_id PK
        text realm
        text action
        text changed_by
        timestamptz changed_at
        jsonb old_row
        jsonb new_row
    }
    users_data {
        bigserial data_id PK
        text realm FK
        bigint id FK
        jsonb payload
        vector embedding
        timestamptz timestamp
    }

    users ||--o{ likes : "from_id"
    posts ||--o{ likes : "to_id"
    users ||--o{ users_audit : "logs changes"
    users ||--o{ users_data : "appends records & embeddings"

🚀 Quick Start Examples

1. Basic Graph Storage & Traversals (asyncpg)

import asyncio
from post_graph import AsyncPostGraph

async def main():
    client = AsyncPostGraph(dsn="postgresql://postgres:postgres@localhost:5432/postgres")
    await client.connect()

    # 1. Create schema tables
    await client.create_vertex_table("users")
    await client.create_vertex_table("posts")
    await client.create_edge_table("likes", from_vertex_table="users", to_vertex_table="posts")

    # 2. Add vertices
    alice = await client.add_vertex("users", realm="realm_a", payload={"name": "Alice"})
    post = await client.add_vertex("posts", realm="realm_a", payload={"title": "Hello PostGraph!"})

    # 3. Add edge connecting Alice -> Post
    edge = await client.add_edge(
        table_name="likes",
        realm="realm_a",
        from_id=alice.id,
        to_id=post.id,
        relation_type="like",
        payload={"reaction": "heart"}
    )

    # 4. Append data history record
    await alice.add_data({"status": "active", "login_count": 1})
    history = await alice.get_data()
    print(f"Alice history count: {len(history)}, last: {history[0].payload}")

    # 5. Object-oriented traversal
    steps = await alice.to("likes")
    for step in steps:
        target_post = step.vertex()
        print(f"Alice liked post: {target_post.payload['title']}")

    await client.close()

asyncio.run(main())

2. pgvector Similarity Search (main, data, or both scopes)

import asyncio
from post_graph import AsyncPostGraph

async def main():
    client = AsyncPostGraph(dsn="postgresql://postgres:postgres@localhost:5432/postgres")
    await client.connect()

    # 1. Create vertex table with 4D embeddings
    await client.create_vertex_table("documents", realm="tech_realm", vector_dim=4)

    # 2. Add vertex with embedding to MAIN table
    doc1 = await client.add_vertex(
        "documents",
        realm="tech_realm",
        payload={"title": "PostgreSQL Overview"},
        embedding=[1.0, 0.0, 0.0, 0.0]
    )

    # 3. Add historical entry with embedding to DATA table
    await doc1.add_data(
        payload={"version": "v2.0", "notes": "Added pgvector support"},
        embedding=[0.0, 1.0, 0.0, 0.0]
    )

    # 4. Vector search on MAIN table
    results_main = await client.vector_search(
        "documents",
        realm="tech_realm",
        query_vector=[0.9, 0.1, 0.0, 0.0],
        top_k=3,
        search_scope="main"
    )
    print("Main Search Hit:", results_main[0][0].payload['title'], "Dist:", results_main[0][1])

    # 5. Vector search on BOTH main & data history tables combined
    results_both = await client.vector_search(
        "documents",
        realm="tech_realm",
        query_vector=[0.05, 0.95, 0.0, 0.0],
        top_k=3,
        search_scope="both"
    )
    print("Combined Scope Best Hit:", results_both[0][0].payload['title'], "Dist:", results_both[0][1])

    await client.close()

asyncio.run(main())

3. Using SQLAlchemyPostGraph

import asyncio
from sqlalchemy.ext.asyncio import create_async_engine
from post_graph import SQLAlchemyPostGraph

async def main():
    engine = create_async_engine("postgresql+asyncpg://postgres:postgres@localhost:5432/postgres")
    client = SQLAlchemyPostGraph(engine)

    # DDL & CRUD methods share identical signatures
    await client.create_vertex_table("users")
    await client.create_edge_table("follows", from_vertex_table="users", to_vertex_table="users")

    v1 = await client.add_vertex("users", realm="default", payload={"name": "Bob"})
    v2 = await client.add_vertex("users", realm="default", payload={"name": "Charlie"})
    e = await client.add_edge("follows", realm="default", from_id=v1.id, to_id=v2.id, relation_type="follow")

    await engine.dispose()

asyncio.run(main())

📖 Detailed API Reference

Client Initialization

AsyncPostGraph(
    dsn: Optional[str] = None,
    connection_or_pool: Optional[Union[asyncpg.Pool, asyncpg.Connection]] = None,
    schema_per_realm: bool = False
)

SQLAlchemyPostGraph(
    engine_or_connection: Union[AsyncEngine, AsyncConnection],
    schema_per_realm: bool = False
)

• schema_per_realm: If True, creates physically isolated schema namespaces per realm (e.g. "realm_a"."users"). If False (default), uses single-schema logical multi-tenancy with a realm column partition.

Schema DDL Operations

create_vertex_table

await client.create_vertex_table(
    table_name="users",
    realm=None,
    vector_dim=None  # Optional integer (e.g. 1536, 4096) for pgvector HNSW index
)

Creates:

1. {table_name} main vertex table with (realm, id) primary key.
2. {table_name}_audit shadow audit log table.
3. {table_name}_data append-only history table with foreign key (realm, id) ON DELETE CASCADE.
4. If vector_dim is provided, creates embedding vector(dim) column and HNSW index on both {table_name} and {table_name}_data.

create_edge_table

await client.create_edge_table(
    table_name="likes",               # Optional; defaults to "{from_table}TO{to_table}" if empty
    from_vertex_table="users",
    to_vertex_table="posts",
    realm=None,
    cascade_delete_from=False,       # If True, deleting edge deletes source vertex
    cascade_delete_to=False          # If True, deleting edge deletes target vertex
)

Vector Similarity Search (vector_search)

results = await client.vector_search(
    table_name="documents",
    realm="tech_realm",
    query_vector=[0.1, 0.2, 0.3, 0.4],
    top_k=5,
    distance_metric="cosine",        # "cosine" (<=>), "l2" (<->), or "inner_product" (<#>)
    search_data_table=False,         # Deprecated alias; set search_scope="data" instead
    search_scope="main"              # "main", "data", or "both"
)

• search_scope="main": Searches vector column on main vertex table {table_name}.
• search_scope="data": Searches vector column on associated data table {table_name}_data joined with main table {table_name}.
• search_scope="both": Performs a unified CTE vector distance query across main and associated data tables, returning distinct matching Vertex objects sorted by minimum distance.

Vertex Operations

add_vertex

vertex = await client.add_vertex(
    table_name="users",
    realm="tenant_1",
    vertex_id=None,                 # Optional autogenerated BIGSERIAL ID
    payload={"name": "Alice"},       # Dict payload stored as JSONB
    embedding=[0.1, 0.2, 0.3, 0.4],  # Optional vector embedding
    user_id="admin_1"               # Optional user attribution for audit log
)

get_vertex

vertex = await client.get_vertex(table_name="users", realm="tenant_1", vertex_id=1)
# Seamlessly accepts integer ID, FQID ("tenant_1/users/1"), or UUID string ("550e8400-e29b-41d4-a716-446655440000")

get_vertex_by_uuid

vertex = await client.get_vertex_by_uuid(
    table_name="users",
    realm="tenant_1",
    uuid="550e8400-e29b-41d4-a716-446655440000"
)

Performs fast $O(1)$ indexed lookup on the automatically assigned uuid column.

get_edge_by_uuid

edge = await client.get_edge_by_uuid(
    table_name="likes",
    realm="tenant_1",
    uuid="463bf3f0-33f7-47cf-9fa9-46ab83035033"
)

Performs fast indexed lookup for edge records by UUID.

update_vertex

updated = await client.update_vertex(
    table_name="users",
    realm="tenant_1",
    vertex_id=1,
    payload={"name": "Alice Smith"},
    user_id="admin_1"
)

delete_vertex

deleted = await client.delete_vertex(table_name="users", realm="tenant_1", vertex_id=1, user_id="admin_1")

Automatically cascade deletes all connected edges in the database.

Edge Operations

add_edge

edge = await client.add_edge(
    table_name="follows",
    realm="tenant_1",
    edge_id=None,                   # Optional autogenerated BIGSERIAL ID
    from_id=1,
    to_id=2,
    relation_type="friend",
    payload={"since": "2026-01-01"},
    user_id="admin_1",
    check_cycle=True                # Raises CyclicReferenceError if edge creates a cycle
)

delete_edge

deleted = await client.delete_edge(table_name="follows", realm="tenant_1", edge_id=100, user_id="admin_1")

Append-Only Data History ({table_name}_data)

Vertices and edges support appending timestamped historical data snapshots with optional vector embeddings.

Client Data APIs

record = await client.add_vertex_data(
    table_name="users",
    realm="realm_a",
    vertex_id=1,
    payload={"status": "active", "login_count": 10},
    timestamp=None,                 # Optional; defaults to CURRENT_TIMESTAMP
    embedding=[0.1, 0.2, 0.3, 0.4]   # Optional vector embedding
)

history = await client.get_vertex_data(table_name="users", realm="realm_a", vertex_id=1, limit=5)

Model Instance Methods

rec = await vertex.add_data({"status": "active"}, embedding=[0.1, 0.2, 0.3, 0.4])
history = await vertex.get_data(limit=10)

Multi-Tenant Realm Deletion

rows_deleted = await client.delete_realm(realm="tenant_to_remove")

Completely removes all vertices, edges, audit logs, and data history records belonging to the target realm across all tables.

Graph Traversals & Shortest Path (CTEs)

traverse

paths = await client.traverse(
    realm="realm_a",
    start_table="users",
    start_id=1,
    edge_tables=["follows", "likes"],
    max_depth=3
)
for p in paths:
    print(f"Depth: {p['depth']} | Path: {' -> '.join(p['path'])}")

shortest_path

sp = await client.shortest_path(
    realm="realm_a",
    start_table="users",
    start_id=1,
    target_table="posts",
    target_id=42,
    edge_tables=["follows", "likes"],
    max_depth=5
)
if sp:
    print(f"Shortest path found at depth {sp['depth']}: {sp['path']}")

📄 License

This project is licensed under the MIT License.

Developed by Chandan Rajah (chandan.rajah@gmail.com).