Building AI Knowledge Graph Using Graphiti & Neo4j

Graphiti is a compact yet powerful Python library that converts raw text or JSON into an AI Knowledge Graph—a structured store of facts that acts as AI Agent Memory. Below is the exact workflow I used to load FutureSmart AI data into Neo4j, explore the resulting AI Graph Memory, and run hybrid (semantic + keyword) searches.

---

1. Why Graphiti for AI Agent Memory?

• Real‑time inserts – Add new facts without bulk re‑processing.

• LLM‑powered parsing – Entity extraction, relationship mapping, summaries, and embeddings handled automatically.

• Neo4j under the hood – First‑class graph database with Cypher, indexes, and visualization tools.

• Perfect for AI Agent Memory – Your agent can recall structured facts on demand.

---

2. Quick Refresher: Knowledge‑Graph Basics

“Pradip Nichite founded FutureSmart AI.”

Graphiti stores each triple as Neo4j nodes (Entity) and edges (RELATES_TO), plus an Episode node that records the original text—together forming your AI Graph Memory.

---

3. Setup

pip install graphiti-core  # build your AI Knowledge Graph in minutes

import os
from google.colab import userdata

os.environ["OPENAI_API_KEY"] = userdata.get("OPENAI_API_KEY")
neo4j_uri      = "neo4j+s://<your‑instance>.databases.neo4j.io"
neo4j_user     = "neo4j"
neo4j_password = userdata.get("NEO4J_PASSWORD")

---

4. Initialise Graphiti (clean slate)

from graphiti_core import Graphiti
from graphiti_core.utils.maintenance.graph_data_operations import clear_data

graphiti = Graphiti(neo4j_uri, neo4j_user, neo4j_password)
await clear_data(graphiti.driver)              # optional for a fresh start
await graphiti.build_indices_and_constraints() # creates indexes once

---

5. Insert Episodes into Your AI Knowledge Graph

episodes = [
    {
        "name": "About Me",
        "content": "Hi, I'm Pradip Nichite. I am the founder and CEO of FutureSmart AI.",
        "type": EpisodeType.text,
        "description": "intro"
    },
    {
        "name": "About FutureSmart AI",
        "content": "FutureSmart AI builds custom AI solutions for clients.",
        "type": EpisodeType.text,
        "description": "company overview"
    }
]

await add_data(episodes)  # helper loops and calls graphiti.add_episode()

Under‑the‑Hood Workflow

1. Entity extraction (people, orgs, products)

2. Relationship extraction (founded, associated_with, …)

3. Deduplication of entities

4. Summary + embedding generation

5. Graph write → Neo4j (your AI Graph Memory)

All steps run via LLM calls you can inspect in your OpenAI usage logs.

---

6. Visual Exploration of AI Graph Memory

• Blue = Episode nodes (source text)

• Brown = Entity nodes (deduplicated concepts)

• Edges labelled MENTIONS & RELATES_TO connect everything

Run MATCH (n) RETURN n in Neo4j Browser to explore your AI Knowledge Graph. Click any node to view its summary and vector embedding.

---

7. CSV Export at a Glance

Sample from node‑export.csv:

---

8. Query Your AI Agent Memory (Hybrid Search)

query   = "What products FutureSmart has"
results = await graphiti.search(query=query, num_results=3)
print_result(results)

Search Results:
UUID: dae5cb4d-adb2-401f-991f-aa5259127245
Fact: FutureSmart Agent is a product of FutureSmart AI
Valid from: 2025-08-01 07:54:17+00:00
---
UUID: afc8af04-0af2-4fd8-940a-91cf93da6943
Fact: https://agent.futuresmart.ai/ is the website of FutureSmart Agent
Valid from: 2025-08-01 07:54:17+00:00
---
UUID: c6d353e8-6fb1-4263-a0e9-4df21ea5805f
Fact: FutureSmart AI is the company behind AI Demos
Valid from: 2025-08-01 07:54:08+00:00

Graphiti blends embedding similarity with BM25 keywords to surface relevant facts for downstream agents.

---

9. Practical Uses of an AI Graph Memory

• AI Agent Memory for chatbots and assistants

• Enterprise AI Knowledge Graph for internal search

• Neo4j Bloom dashboards for leadership teams

• Light‑weight RAG pipelines powered by graph facts

---

10. What’s Next?

• Integrate Graphiti memory into LangGraph / LangChain agents

• Combine vector store retrieval with graph queries

• Package everything in FastAPI + Streamlit for a live demo

---

Prefer video over text?

Watch the full walkthrough here

---

Need a Custom AI Agent or Graph Solution?

FutureSmart AI helps companies turn raw data into production-ready AI Knowledge Graphs and AI Agent Memory systems.

• End-to-end Graphiti + Neo4j pipelines

• LangChain / LangGraph agent integrations

• Scalable FastAPI & Streamlit front-ends

• Secure, cloud-native deployment

Want proof we deliver? 👉 See our real-world results: https://futuresmart.ai/case-studies

Book a free consult: contact@futuresmart.ai

---

Questions?

Drop a comment on YouTube or email contact@futuresmart.ai.