Memory SDK

A Python SDK for storing structured memory objects in Supabase using OpenAI's API.

Installation

# Clone the repository
git clone https://github.com/deadcow-labs/human-memory.git
cd human-memory

# Install dependencies
pip install -r requirements.txt

Requirements

Create a requirements.txt file with the following dependencies:

openai>=1.0.0
supabase>=2.0.0
python-dotenv>=1.0.0

Configuration

The SDK requires the following environment variables:

• OPENAI_API_KEY: Your OpenAI API key
• SUPABASE_URL: Your Supabase project URL
• SUPABASE_KEY: Your Supabase API key

You can set these in your environment or use a .env file with python-dotenv.

Supabase Setup

Create a memories table in your Supabase project with the following schema:

create table
  public.memories (
    id uuid primary key,
    user_id text not null,
    created_at timestamp with time zone not null,
    content text not null,
    reflection text not null,
    embedding vector(1536) not null,
    emotional_tone text not null,
    location jsonb not null,
    tags text[] not null
  );

Enable RLS (Row Level Security) on your table and create appropriate policies.

Usage

from memory_sdk import MemorySDK

# Initialize the SDK with a user ID
user_id = "user_123"
sdk = MemorySDK(user_id)

# Save a memory from a raw message
raw_message = "Today I made significant progress on my project and feel optimistic about finishing it soon."
memory = sdk.save(raw_message)

# The memory object contains all structured data:
print(f"Content: {memory.content}")
print(f"Reflection: {memory.reflection}")
print(f"Emotional tone: {memory.emotional_tone}")

See example.py for a complete usage example.

Memory Structure

Each memory includes:

• id: UUID
• user_id: provided by the SDK user
• created_at: UTC timestamp
• content: short summary of the message
• reflection: insight about the user from the message
• embedding: vector from OpenAI embedding API
• emotional_tone: e.g. "hopeful", "anxious"
• location: dict with type and name
• tags: list of string topics or themes

License

MIT