Case Study

FlockX & Zep AI

Enabling Real-World Connections through Advanced Memory Technology

FlockX, a startup based in Grand Rapids, Michigan, is tackling the challenge of transforming online interactions into meaningful, real-world connections through their AI-powered social platform. To enhance their platform's ability to provide personalized and context-aware user experiences, FlockX has integrated Zep AI's AI agent memory technology. Zep AI's memory layer equips FlockX's AI sidekicks with long-term memory, automatically storing and maintaining user preferences, social connections, and other user attributes. Zep intelligently grounds agents in the correct context, enabling FlockX to deliver more engaging and relevant interactions and ultimately foster deeper social connections within local communities.

Challenge

Enhancing AI Sidekicks with Persistent Memory

FlockX's mission is to address the growing disconnect between online interactions and meaningful, real-world connections. Research has shown that chronic loneliness can have severe health implications, increasing the risk of dementia by 50% and heart disease by 29%. FlockX aims to combat this issue by building 'flocks' - groups of people and their AI sidekicks - that work together to foster authentic relationships in local communities. These AI sidekicks act as personal guides, offering activity recommendations, facilitating event planning, and fostering connections between users with shared interests. To achieve this vision, FlockX required a memory solution that could:

Send Zep a stream of chat messages & business data

Zep intelligently fuses data on a Knowledge Graph

Provide your agent accurate and relevant user facts with a single API call

Solution

Integrating Zep's Long-Term Memory Layer

Zep AI, a temporal knowledge graph-based memory layer for AI assistants, was the ideal solution to address FlockX's needs. By integrating Zep AI into their platform, FlockX gained access to:

Intelligent persistence of conversation and business data: Zep automatically, and intelligently builds a knowledge graph, an ideal data structure for modeling the complex world of FlockX where user preferences are learned and change over time, growing social networks, and many disparate social events for users to participate in.

Simple and accurate context retrieval: Zep’s high-level memory APIs make it simple to retrieve highly relevant and comprehensive context data for inclusion in an agent’s prompt. Zep also provides lower-level APIs to search the Zep graph, with sophisticated semantic and graph-based search approaches offered. These can be used by developers directly, or provided to an agent as tools.

Privacy and Security: Zep AI prioritizes data privacy, ensuring that sensitive user information is handled securely. This focus aligns with FlockX's commitment to user data protection, building trust and transparency within their platform.

Implementation

Empowering AI Sidekicks with Context & Group Graphs

As users interact with FlockX, their actions, preferences, and conversations are used to construct personalized knowledge graphs within Zep AI. These graphs represent each user's unique social world, including their connections, interests, and past experiences.

In addition to individual user graphs, FlockX leverages Zep AI's Group Graphs to capture shared context and facilitate interactions within communities. Group graphs allow FlockX to create rich, shared knowledge representations that enhance the social experience and enable more meaningful connections.

Creating Group Graphs

Creating Group Graphs: To create a group graph, FlockX uses the graph.add() method, as demonstrated in the following code snippet:

 
# Example of adding structured JSON data to a group graph for FlockX's NBA partner

from zep_cloud.client import AsyncZep
import asyncio  

async def add_group_data(): 
     zep_client = AsyncZep(api_key="your_api_key")  
     json_data = { 
          "event_type": "watch_party",
          "event_details": { 
               "title": "Grand Rapids Gold Watch Party", 
               "date": "2025-02-15", 
               "location": "Downtown Sports Bar", 
               "address": "123 Main St, Grand Rapids, MI", 
               "start_time": "18:00", 
               "end_time": "21:00" 
          }, 
          "shared_interests": ["basketball", "NBA G League", "Grand Rapids Gold"], 
          "organizer": { 
               "name": "Buckets AI Sidekick", 
               "contact_email": "[email protected]"
          }, 
          "participants": [ 
               {"name": "Alice Johnson", "status": "confirmed"}, 
               {"name": "Mark Simmons", "status": "invited"}, 
               {"name": "Taylor Nguyen", "status": "confirmed"} 
          ] 
     }  

     group_id = "grand_rapids_gold_fan_group"  

     # Adding data to the group graph 
     await zep_client.graph.add( 
          data=json_data, 
          group_id=group_id, 
          type="json" 
     )

 
import { AsyncZep } from "zep-cloud-client";

async function addGroupData() {

const zepClient = new AsyncZep({ apiKey: "your_api_key" });

const jsonData = {

event_type: "watch_party",

event_details: {

title: "Grand Rapids Gold Watch Party",

date: "2025-02-15",

location: "Downtown Sports Bar",

address: "123 Main St, Grand Rapids, MI",

start_time: "18:00",

end_time: "21:00"

},

shared_interests: ["basketball", "NBA G League", "Grand Rapids Gold"],

organizer: {

name: "Buckets AI Sidekick",

contact_email: "[email protected]"

},

participants: [

{ name: "Alice Johnson", status: "confirmed" },

{ name: "Mark Simmons", status: "invited" },

{ name: "Taylor Nguyen", status: "confirmed" }

]

};

const groupId = "grand_rapids_gold_fan_group";

// Adding data to the group graph

await zepClient.graph.add({

data: jsonData,

groupId: groupId,

type: "json"

});

console.log("Group data added successfully");

}

addGroupData().catch(console.error);

 
package main

import (
	"bytes"
	"encoding/json"
	"fmt"
	"net/http"
)

func addGroupData() {
	apiKey := "your_api_key"
	url := "https://api.zep.cloud/v1/graph/add"

	jsonData := map[string]interface{}{
		"event_type": "watch_party",
		"event_details": map[string]string{
			"title":      "Grand Rapids Gold Watch Party",
			"date":       "2025-02-15",
			"location":   "Downtown Sports Bar",
			"address":    "123 Main St, Grand Rapids, MI",
			"start_time": "18:00",
			"end_time":   "21:00",
		},
		"shared_interests": []string{"basketball", "NBA G League", "Grand Rapids Gold"},
		"organizer": map[string]string{
			"name":          "Buckets AI Sidekick",
			"contact_email": "[email protected]",
		},
		"participants": []map[string]string{
			{"name": "Alice Johnson", "status": "confirmed"},
			{"name": "Mark Simmons", "status": "invited"},
			{"name": "Taylor Nguyen", "status": "confirmed"},
		},
		"group_id": "grand_rapids_gold_fan_group",
		"type":     "json",
	}

	jsonValue, err := json.Marshal(jsonData)
	if err != nil {
		fmt.Println("Error marshaling JSON:", err)
		return
	}

	req, err := http.NewRequest("POST", url, bytes.NewBuffer(jsonValue))
	if err != nil {
		fmt.Println("Error creating request:", err)
		return
	}
	req.Header.Set("Content-Type", "application/json")
	req.Header.Set("Authorization", "Bearer "+apiKey)

	client := &http.Client{}
	resp, err := client.Do(req)
	if err != nil {
		fmt.Println("Error making request:", err)
		return
	}
	defer resp.Body.Close()

	fmt.Println("Response Status:", resp.Status)
}

func main() {
	addGroupData()
}

Here, json_data represents the structured data that FlockX wants to add to the group graph, such as community information, shared interests, or event details. The group_id parameter identifies the specific group graph to which the data should be added.

Next, FlockX's AI sidekicks were upgraded to leverage Zep AI's memory capabilities. During conversations, the sidekicks utilize Zep AI to retrieve relevant facts and context from the user's knowledge graph, allowing them to provide informed responses and recommendations.

Results

Enhanced User Experience and Social Engagement

The integration of Zep AI has significantly enhanced the FlockX platform, leading to:

Personalized Recommendations:

Engaging Conversations

AI sidekicks can now maintain context throughout conversations, remembering past discussions and referencing previous user input. This conversational depth creates more natural and engaging interactions, fostering stronger connections between users and their AI companions.

Improved Social Planning

Flocks, FlockX's private group chat feature, benefit greatly from Zep AI's memory capabilities. AI sidekicks can assist with group planning, suggesting activities that align with the collective interests of the flock, and even generating itineraries based on shared preferences.

Looking ahead

By continuously refining the understanding of user preferences and behaviors, the partnership aims to create a platform that seamlessly connects individuals with shared passions and drives meaningful, real-world interactions.
‍
As part of its ongoing evolution, FlockX is also developing a no-code version of its platform. This new feature will enable anyone to create a knowledge agent that seamlessly connects to Zep, expanding its accessibility beyond community-focused use cases. By making AI-powered tools more accessible, FlockX hopes to empower more users to build engaging experiences that foster stronger connections and social interactions.

Citations