mem0 Demo Project - Streamlit Chat

This project demonstrates how to create a learning chatbot using mem0 library for AI memory management, integrating it with Google Gemini, Supabase (vector store), and Neo4j (graph store) within a Streamlit chat application.

This is mainly a playground to investigate and learn and should be viewed as such.

Features

• ✅ Streaming chat interface powered by Streamlit.
• ✅ Conversation processing and memory management using mem0.
• ✅ LLM integration with Google Gemini for chat generation and memory management.
• ✅ Vector memory storage via Supabase.
• ✅ Graph memory storage via Neo4j.
• ✅ Persistence of memories for each user across sessions.
• ✅ User authentication using Supabase Auth.
• ✅ Display prompt, Response and Total tokens count for the current prompt and conversation.
• ✅ Sidebar displaying the LLM model in use and an option to clear the conversation UI.
• ✅ Display Mem0 memory interactions.
• ✅ Implement basic Google Gemini moderation.

(back to top)

Prerequisites

1. uv for Python version management and dependency management

   uv is a fast Python package installer and resolver, written in Rust. To check if uv is already installed, run:

   uv --version

   If uv is not found, you can install it using a standalone installer or through your preferred package manager. Make sure uv --version returns a version number before proceeding. For instance:

   uv --version
   uv 0.8.3 (7e78f54e7 2025-07-24)
   

(back to top)

Setup Instructions

1. Clone the Repository:

   git clone https://github.com/ytkaczyk/gemini-mem0-streamlit.git
   cd gemini-mem0-streamlit

2. Install Dependencies:

   • Install all required packages using uv sync:

     uv sync

   • (This installs all necessary libraries including mem0ai, streamlit, python-dotenv, google-generativeai, and their dependencies specified in the uv.lock file.)

3. Get a Google Gemini API Key:

   • Obtain a Google Gemini API key from Google AI Studio.

4. Setup Supabase:

   • Sign up for a free Supabase account at https://supabase.com/.
   • Create a new project.
   • Navigate to "Project Settings" -> "API" to find your SUPABASE_URL and SUPABASE_ANON_KEY.
   • Go to "Database" -> "Connection String" to get your SUPABASE_CONNECTION_STRING. Ensure you use the "URI" format.
   • Enable the vector extension in your Supabase database:
     • Go to "Database" -> "Extensions" and search for "vector".
     • Toggle it on.
   • Configure the Authentication (optional)
     • For the purpose of the demo and to simplify account creation setup and allow using any email address (e.g. test@example.com), it is convenient to disable email confirmation following account creation.

5. Setup Neo4j:

   • Sign up for a free Neo4j AuraDB account at https://neo4j.com/cloud/aura/.
   • Create a new Neo4j instance (a free tier is available).
   • Once your instance is created, download the connection details (usually a .txt file). This file will contain your NEO4J_URI, NEO4J_USERNAME, and NEO4J_PASSWORD.

6. Configure Environment Variables:

   • Copy .example.env to .env in the project root directory (where app.py is located).
   • Add the credentials from the previous steps to the .env file. Never commit this file to Git.

     # Gemini config
     # Obtain a key at https://aistudio.google.com/app/apikey
     GOOGLE_API_KEY="<your api key>"
     # llm and embedding model
     # Model list at https://ai.google.dev/gemini-api/docs/models
     LLM_MODEL="gemini-2.0-flash"
     # Embedding model list at https://ai.google.dev/gemini-api/docs/models#gemini-embedding and https://ai.google.dev/gemini-api/docs/models#text-embedding
     EMBEDDING_MODEL="models/gemini-embedding-001"
     EMBEDDING_MODEL_DIMS=1536
     
     # Supabase config
     # Sign up for a free account at https://supabase.com/ or self host
     SUPABASE_URL="https://<supabase project ID>.supabase.co"
     SUPABASE_ANON_KEY="<your anon key>"
     SUPABASE_TABLE_NAME="mem0_memories"
     SUPABASE_CONNECTION_STRING="<database connection string>"
     
     # Neo4j config
     # Sign up for a free Neo4j AuraDB account at https://neo4j.com/product/auradb/ or self host. 
     NEO4J_URI="neo4j+s://<neo4j aura project ID>.databases.neo4j.io" 
     NEO4J_USERNAME="<usually neo4j>"
     NEO4J_PASSWORD="<your neo4j password>"

   • Ensure your Supabase database has the vector extension enabled.
   • Ensure your Neo4j instance is running and accessible.

7. Run the Application:

   • Execute the following command in your terminal (with the virtual environment activated):

     streamlit run app.py

   • The application should open in your default web browser.
   • Create a user by entering an email address, password and clicking on Sign Up. If account creation email confirmation was disabled in Supabase, the account is immediately available and can be used for logging in.

(back to top)

AI coding assistance with Cline

Cline AI for VSCode was used to generate the application. Cline AI is an open-source coding agent that can understand entire codebases, plan complex changes, and execute multi-step tasks. In order to provide strong guidance and predictable steps to the coding agent, the application was generated using the following steps:

• Create .clinerules file to provide overall rules to Cline. This file is somewhat generic and can be reused across projects. See documentation. The Github Copilot documentation also provides good guidance on creating system-wide, project and personal level rules.
• Create the planning.md and task.md files. This was done using the following prompt:

I am planning to create a project with python for demonstrating the mem0 library (https://docs.mem0.ai/overview). I want the demo to use:
- Supabase as a vector database (https://docs.mem0.ai/components/vectordbs/dbs/supabase)
- neo4j as a graph database (https://docs.mem0.ai/open-source/graph_memory/overview)
- gemini as LLM (https://docs.mem0.ai/components/llms/models/gemini)
- Provide a chat based frontend using streamlit (https://streamlit.io/)
You can perform grounded searches to learn more about the technology. 
Then create a PLANNING.md and TASK.md files.

PLANNING.md - high level direction, scope, tech, etc.
TASK.md - initial tasks to knock out for the project

• Review and edit the planning.md and task.md files.
• Ask Cline to implement the code by following the task sequence in task.md.
• Add new features by asking Cline to add tasks to task.md and then implementing them.

Points of interest

Showing memory information

Toggling the Show Memory Information under the Conversation section is helpful to see which memories are being retrieved and provided as extra context to the LLM prior to answering a prompt as well as how memories are add, deleted or updated.

The following conversation shows how a memory can be updated:

View user's memory

Navigate to the Memory to view the stored memories for the current user, both in the vector and graph data stores.

• Vector memories

• Graph memories

(back to top)

How it Works

See PLANNING.md for more architectural details and TASK.md for development status.

High level architecture

flowchart TD;
  Streamlit["Streamlit UI (Chat Interface)"];
  PythonApp["Python App (app.py)"];
  Memory["mem0 (Memory Client)"];
  Gemini["Gemini LLM (Processing & Gen)"];
  Supabase["Supabase (Vector Store)"];
  Neo4j["Neo4j (Graph Store)"];

  Streamlit --> PythonApp;
  PythonApp <-->|Stores & Retrieves data| Memory; 
  Memory -->|Interacts With| Gemini;
  Memory -->|Stores Data In| Supabase;
  Memory -->|Links Data In| Neo4j;
  PythonApp --> |Interacts With| Gemini;

Loading

(back to top)