Skip to main content

Introduction

Create Context Graph is an interactive CLI scaffolding tool that generates complete, domain-specific context graph applications. Think of it as create-next-app, but for AI agents backed by graph memory.

Given a domain (like healthcare, financial services, or wildlife management) and an agent framework, it generates a full-stack application: a FastAPI backend with a configured AI agent, a Next.js + Chakra UI frontend with NVL graph visualization, a Neo4j schema with synthetic data, and domain-specific tools that let the agent query and reason over your knowledge graph.

The generated app's three-panel layout: chat interface, graph visualization, and document browser

What is POLE+O?

The POLE+O entity model is the foundation for all context graphs: Person, Organization, Location, Event, plus Object. Every domain ontology inherits these five base types and adds domain-specific subtypes. See How Domain Ontologies Work for details.

Key Features​

  • 22 built-in domains -- Healthcare, financial services, real estate, manufacturing, scientific research, software engineering, and more. Each ships with a complete ontology, agent tools, demo scenarios, and fixture data.
  • 8 agent frameworks -- PydanticAI, Claude Agent SDK, OpenAI Agents SDK, LangGraph, CrewAI, Strands, Google ADK, and Anthropic Tools.
  • Multi-turn conversations -- Every agent uses neo4j-agent-memory for conversation persistence with automatic entity extraction and preference detection.
  • Graph-native AI agents -- Cypher-powered tools for querying entities, relationships, and decision traces. Tool calls stream in real-time with live progress indicators.
  • Streaming chat -- Token-by-token responses via Server-Sent Events. Tool calls appear as a live timeline with spinner indicators. Graph visualization updates incrementally after each tool completes.
  • Interactive graph visualization -- NVL-powered graph explorer with entity detail panel, document browser with template filtering, and decision trace viewer.
  • Rich demo data -- LLM-generated fixture data per domain: 80-90 entities, 25+ professional documents, and 3-5 multi-step decision traces. Loaded via make seed.
  • Flexible Neo4j setup -- Neo4j Aura (free cloud), @johnymontana/neo4j-local, Docker Compose, or any existing instance.
  • 12 SaaS data connectors -- GitHub (github), Slack (slack), Jira (jira), Notion (notion), Gmail (gmail), Google Calendar (gcal), Salesforce (salesforce), Linear (linear), Google Workspace (google-workspace), Claude Code (claude-code), Claude AI (claude-ai), and ChatGPT (chatgpt). Use the ID in parentheses with --connector.
  • Custom domains -- Describe your domain in natural language to generate a complete ontology, or write your own YAML definition.
  • MCP server for Claude Desktop -- Optionally generate an MCP server config so Claude Desktop queries the same knowledge graph as your web app.

Quick Install​

No installation required. Run directly with uvx (Python) or npx (Node.js):

# Python (recommended)
uvx create-context-graph

# Node.js
npx create-context-graph

See the Quick Start for a complete walkthrough, or skip the wizard with flags:

uvx create-context-graph my-app --domain healthcare --framework pydanticai --demo-data

See All Available Domains​

uvx create-context-graph --list-domains

Architecture​

Architecture: generation pipeline (CLI → Jinja2 → backend + frontend + data) and runtime (frontend ↔ backend ↔ Neo4j)

The top half shows generation: the CLI reads a domain ontology YAML and renders Jinja2 templates into a complete project (FastAPI backend, Next.js frontend, Cypher schema + fixture data). The bottom half shows the running application: the frontend streams chat responses via SSE, the backend agent executes Cypher tool calls against Neo4j, and the graph visualization updates incrementally as each tool completes.

Reading Guide​

Choose your path based on what you want to do:

What's Next​