Graphiti Integration

Overview

Graphiti is an open-source framework that transforms text, JSON, and chat content into queryable knowledge graphs using language models and graph databases. It excels at:

Semantic indexing and retrieval
Contextual search across structured/unstructured data
Building long-term memory for AI agents

Why use Graphiti with Relax AI? Combine Neo4j’s graph capabilities with Relax’s powerful LLMs for context-aware applications.

Key Features

Multi-format Ingestion: Process text, JSON, and chat messages
Hybrid Search: Combine vector similarity and graph traversal
Relax AI Integration: Use any Relax AI model for embeddings and reasoning
Neo4j Backend: Lever industry-leading graph database
Memory API: Create persistent knowledge for AI agents

Use Cases

Building corporate knowledge graphs from documents and chats
Creating contextual memory systems for AI copilots
Real-time reasoning over dynamic data streams
Customer support history analysis and retrieval
Integrating Slack/PDF data into queryable knowledge

Quick Setup

Install Graphiti Core
Terminal window
```
pip install graphiti-core openai
```
Set up Neo4j database

Graphiti requires Neo4j version 5.26 or higher. Either download and install Neo4j Desktop. Create a project, add a new database (Local DBMS), set password for user neo4j, and start the database or start a docker container with the following command:
Terminal window
```
docker run \
 --name neo4j-graphiti \
 -p 7474:7474 -p 7687:7687 \
 -e NEO4J_AUTH=neo4j/your_password \
 -e NEO4J_PLUGINS='["apoc"]' \
 neo4j:5.26-community
```
This exposes the HTTP (7474) and Bolt (7687) ports with neo4j/your_password credentials.

Set required environment variables

Can be in .env file:

export NEO4J_URI="bolt://localhost:7687"
export NEO4J_USER="neo4j"
export NEO4J_PASSWORD="your-password"

export OPENAI_API_KEY="rak_..." # Replace with your Relax AI API key
export OPENAI_BASE_URL="https://api.relax.ai/v1"

Initialize the Neo4j database schema

Run the below sample script to create the schema:

import os
import asyncio
from graphiti_core import Graphiti

async def main():
    graphiti = Graphiti(
        os.getenv("NEO4J_URI"),
        os.getenv("NEO4J_USER"),
        os.getenv("NEO4J_PASSWORD")
    )
    await graphiti.build_indices_and_constraints(delete_existing=True)
    print("✅ Neo4j schema initialized")
    await graphiti.close()

if __name__ == "__main__":
    asyncio.run(main())

Configure Graphiti with RelaxAI client

Now configure Graphiti with RelaxAI client along with a sample python example below.

import os, asyncio, json
from graphiti_core import Graphiti
from graphiti_core.llm_client import LLMConfig, OpenAIClient
from graphiti_core.embedder.openai import OpenAIEmbedder, OpenAIEmbedderConfig
from graphiti_core.nodes import EpisodeType
from datetime import datetime, timezone

# LLM & Embedder setup
llm_config = LLMConfig(model="GLM-46", small_model="Llama-4-Maverick-17B-128E")
llm_client = OpenAIClient(llm_config, None)
embedder = OpenAIEmbedder(OpenAIEmbedderConfig(embedding_model="Mistral-7b-embedding"), None)

async def main():
    graphiti = Graphiti(
        os.getenv("NEO4J_URI"),
        os.getenv("NEO4J_USER"),
        os.getenv("NEO4J_PASSWORD"),
        llm_client=llm_client,
        embedder=embedder
    )

    await graphiti.build_indices_and_constraints(delete_existing=True)
    print("✅ Schema built")

    # episodes
    episodes = [
        {
            "name": "acquisition-news",
            "body": (
                "In April 2025, Acme Corp acquired BetaTech for $25M. "
                "Jane Doe, VP of Strategy at Acme, will lead integration."
            ),
            "source": EpisodeType.text,
            "description": "Corporate acquisition announcement",
            "time": datetime(2025, 4, 1, tzinfo=timezone.utc)
        },
        {
            "name": "customer-support-chat",
            "body": (
                "Customer: I've been using your mobile app since last year, and the notifications aren't working.\n"
                "Support: Hi! We rolled out a fix in March 2025. Please update the app."
            ),
            "source": EpisodeType.message,
            "description": "Customer support chat",
            "time": datetime(2025, 4, 5, tzinfo=timezone.utc)
        },
        {
            "name": "product-release",
            "body": json.dumps({
                "product_id": "P-100",
                "name": "EcoSmart Thermostat",
                "manufacturer": "HomeTech Inc",
                "price": 199.99,
                "release_date": "2025-03-20T10:00:00Z",
                "features": ["Remote control", "Energy saving mode", "Voice integration"]
            }),
            "source": EpisodeType.json,
            "description": "Product catalog entry",
            "time": datetime(2025, 3, 20, tzinfo=timezone.utc)
        }
    ]

    for ep in episodes:
        await graphiti.add_episode(
            name=ep["name"],
            episode_body=ep["body"],
            source=ep["source"],
            source_description=ep["description"],
            reference_time=ep["time"]
        )
        print(f"✅ Added episode: {ep['name']}")

    # Sample search
    results = await graphiti.search("Who did Acme Corp acquire?")
    print("Search results:", results)

    await graphiti.close()

if __name__ == "__main__":
    asyncio.run(main())

Advanced Usage

Namespaces (group_id)

You can assign a group_id when initializing Graphiti to separate memory graphs by application, tenant, or agent.

graphiti = Graphiti(
    uri, user, password,
    llm_client=llm_client,
    embedder=embedder,
    group_id="team-42"
)

Hybrid Search

By default, graphiti.search() uses vector similarity + graph traversal. You can override search strategies, weights, or filter by time, entity type, or tag.

Graph Queries (Cypher)

Use Neo4j Browser or Bloom to run the following queries:

// Get all people involved in launches
MATCH (e:Event)<-[:INVOLVED_IN]-(p:Entity)
WHERE e.name CONTAINS 'launch'
RETURN p.name, e.date

// Find related support tickets
MATCH (t:Ticket)-[r:RELATED_TO]->(p:Product)
RETURN t, p, r

You should see graph nodes representing entities, events, and relationships in the Neo4j Browser.

Graphiti Neo4j Browser

Troubleshooting

Connection Issues: Verify Neo4j is running and credentials are correct
Schema Errors: Run build_indices_and_constraints(delete_existing=True)
Model Compatibility: Ensure Relax AI model names match available models