django-graphify

Zero-LLM knowledge graph for Django projects.
Inspect models, views, URLs, signals, serializers, admin classes, middleware and settings — all fully offline.

Features

Installation

pip install django-graphify

Add to INSTALLED_APPS:

INSTALLED_APPS = [
    ...
    "django_graphify",
]

Quick start

# Full project graph — all formats
python manage.py graphify

# Single app
python manage.py graphify --app orders

# Custom output dir
python manage.py graphify --out ./docs/graph

# Only json + html
python manage.py graphify --format json html

# Cleaner graph — skip field-level nodes
python manage.py graphify --no-fields

# Only EXTRACTED edges (skip inferred queryset links)
python manage.py graphify --skip-inferred

Output lands in ./graphify-out/ by default:

graphify-out/
├── graph.html         ← interactive force-directed graph (open in browser)
├── graph.json         ← queryable node/edge data
├── GRAPH_REPORT.md    ← god nodes, URL chains, signal wiring, orphans
├── graph.cypher       ← Neo4j import
└── graph.mermaid      ← Mermaid flowchart (paste into any renderer)

Command reference

Node types

Edge types

Every edge is tagged EXTRACTED (found in source), INFERRED (confident guess), or AMBIGUOUS (needs review). INFERRED edges carry a confidence score (0–1).

Python API

You can also drive extraction programmatically:

from pathlib import Path
from django_graphify.graph import GraphBuilder
from django_graphify.extractors import ModelExtractor, ViewExtractor
from django_graphify.exporters import JsonExporter, HtmlExporter

builder = GraphBuilder(project_name="myproject")
root = Path("/path/to/project")

for ExtractorClass in [ModelExtractor, ViewExtractor]:
    extractor = ExtractorClass(root)
    nodes, edges = extractor.ast_extract()
    builder.add_nodes(nodes)
    builder.add_edges(edges)
    nodes, edges = extractor.introspect()
    builder.add_nodes(nodes)
    builder.add_edges(edges)

JsonExporter(builder).export(Path("./out"))
HtmlExporter(builder).export(Path("./out"))

# Query
node = builder.get_node("orders.Order")
neighbours = builder.get_neighbours("orders.Order")
path = builder.shortest_path("orders.Order", "users.User")
gods = builder.god_nodes(top_n=5)
orphans = builder.orphan_nodes()

graph.json schema

{
  "meta": {
    "project": "myproject",
    "django_version": "4.2.7",
    "generated_at": "2025-04-29T10:00:00Z",
    "node_count": 142,
    "edge_count": 89
  },
  "nodes": [
    {
      "id": "orders.Order",
      "type": "model",
      "label": "Order",
      "app": "orders",
      "file": "orders/models.py",
      "line": 14,
      "tag": "EXTRACTED",
      "confidence": 1.0,
      "meta": { "db_table": "orders_order", "abstract": false }
    }
  ],
  "edges": [
    {
      "source": "orders.Order",
      "target": "users.User",
      "relation": "fk",
      "tag": "EXTRACTED",
      "confidence": 1.0,
      "meta": { "field": "customer" }
    }
  ]
}

How it works

Pass 1 — AST (always runs):
Uses Python's ast module to walk every .py file. No Django import required. Extracts class names, base classes, field assignments, decorators. Works even if the project can't be fully imported.

Pass 2 — Django introspection (runs when Django is configured):
Uses django.apps.apps.get_models(), django.conf.settings, and DRF's serializer registry to enrich Pass 1 nodes with concrete field types, FK targets, choices, related_name, middleware ordering, and installed app paths. Gracefully skipped if DJANGO_SETTINGS_MODULE isn't set.

Requirements

• Python 3.10+
• Django 3.2+
• networkx >= 3.0
• djangorestframework (optional — DRF extraction auto-detected)

Author

Yogesh Chauhan — AI and Django engineer, India.

GitHub | LinkedIn | Portfolio

License

MIT — see LICENSE.