Octopus Sensing SARA

An empathetic conversational AI companion with persistent memory, user profiling, and multi-LLM support. Part of the Octopus Sensing project ecosystem.

Built with FastAPI, LangChain, and React.

---

Features

• 🧠 Empathetic AI - Emotionally intelligent responses with advanced prompt engineering
• 💾 Persistent Memory - SQLite storage for conversations and user profiles
• 👤 User Profiling - Automatic extraction of preferences and personal information
• 🔌 Multi-LLM Support - OpenAI, Anthropic (Claude), Google Gemini, and Ollama
• 📝 Auto-Summarization - Long-term conversation summaries for context retention
• 🚀 Production-Ready - Docker deployment, type safety, async API
• 🎨 Modern UI - React frontend with shadcn/ui components

---

Quick Start

Prerequisites

• Docker and Docker Compose
• API key for your LLM provider (OpenAI, Anthropic, or Google)

Installation

1. Clone the repository

   git clone https://github.com/octopus-sensing/octopus-sensing-sara.git
   cd octopus-sensing-sara
   

2. Configure environment

   cp .env.example .env
   

3. Add your API key to .env

   # Choose your provider
   LLM_PROVIDER=gemini
   LLM_MODEL=models/gemini-2.5-flash
   GOOGLE_API_KEY=your-api-key-here
   

4. Start the application

   docker compose up --build
   

5. Access the application

   • Frontend: http://localhost
   • Backend API: http://localhost:8000
   • API Docs: http://localhost:8000/docs

---

Configuration

Configure SARA by editing the .env file:

Variable	Description	Example
LLM_PROVIDER	AI provider	openai, anthropic, gemini, ollama
LLM_MODEL	Model name	models/gemini-2.5-flash, gpt-4, claude-3-opus
LLM_TEMPERATURE	Response creativity (0-2)	0.7
LLM_MAX_TOKENS	Max response length	2000
SHORT_TERM_MEMORY_WINDOW	Recent messages to remember	10
LONG_TERM_MEMORY_SUMMARY_THRESHOLD	Messages before summarization	50

LLM Provider Setup

Google Gemini (Free Tier)

LLM_PROVIDER=gemini
LLM_MODEL=models/gemini-2.5-flash
GOOGLE_API_KEY=your-google-api-key

OpenAI

LLM_PROVIDER=openai
LLM_MODEL=gpt-4
OPENAI_API_KEY=sk-your-openai-key

Anthropic Claude

LLM_PROVIDER=anthropic
LLM_MODEL=claude-3-5-sonnet-20241022
ANTHROPIC_API_KEY=sk-ant-your-anthropic-key

---

API Reference

Send Message

curl -X POST http://localhost:8000/chat \
  -H "Content-Type: application/json" \
  -d '{
    "user_id": "user123",
    "message": "Hello, I need help with something."
  }'

Get User Profile

curl http://localhost:8000/user/user123/profile

Get Conversation History

curl http://localhost:8000/conversation/{session_id}

Full API documentation: http://localhost:8000/docs

---

Architecture

┌─────────────────┐
│  React Frontend │
└────────┬────────┘
         │
┌────────▼────────┐
│   FastAPI REST  │
└────────┬────────┘
         │
┌────────▼──────────────────────────┐
│   Conversation Service             │
│   • Process messages               │
│   • Extract user info              │
│   • Manage memory & summarization  │
└────────┬──────────────────────────┘
         │
┌────────▼────────┬──────────┬──────────┐
│ Memory Service  │  User    │   LLM    │
│                 │ Service  │ Factory  │
└────────┬────────┴────┬─────┴────┬─────┘
         │             │          │
┌────────▼─────────────▼──────────▼─────┐
│     SQLite Database (Persistent)       │
│  • Conversations  • Users  • Messages  │
└────────────────────────────────────────┘

---

Project Structure

octopus-sensing-sara/
├── server/
│   ├── octopus_sensing_sara/
│   │   ├── api/              # REST endpoints
│   │   ├── core/             # Config, LLM factory
│   │   ├── prompts/          # Modular prompt system
│   │   │   ├── system_prompt/
│   │   │   ├── summarization_prompt/
│   │   │   └── extraction_prompt/
│   │   ├── models/           # Database & schemas
│   │   ├── services/         # Business logic
│   │   ├── storage/          # Database layer
│   │   ├── main.py           # FastAPI app
│   │   └── run.py            # Entry point
│   ├── tests/
│   └── Dockerfile
│
├── octopus_sensing_sara_ui/
│   ├── src/
│   │   ├── components/       # UI components
│   │   └── pages/            # Page components
│   └── Dockerfile
│
├── docker-compose.yml
├── pyproject.toml              # Poetry configuration & dependencies
└── poetry.lock

---

How It Works

1. Empathetic Conversations

SARA uses advanced prompt engineering to:

• Recognize emotional states from user messages
• Validate feelings before offering solutions
• Adapt tone based on user's emotional context
• Maintain warm, supportive communication

2. Memory System

Short-Term Memory

• Last 10 messages kept in memory
• Provides immediate conversation context
• Cached per session

Long-Term Memory

• All conversations stored in SQLite
• User profiles with extracted facts
• Automatic summarization after 50 messages

3. User Profiling

SARA automatically extracts and remembers:

• Name and personal details
• Preferences and interests
• Conversation history
• Emotional patterns

This data personalizes future interactions.

4. Modular Prompts

Prompts are organized by function:

• system_prompt/ - SARA's core personality
• summarization_prompt/ - Conversation summaries
• extraction_prompt/ - User information extraction

Each uses LangChain templates for consistency.

---

Docker Commands

Start application

docker compose up -d

View logs

docker compose logs -f backend

Restart after changes

docker compose restart backend

Stop application

docker compose down

Clean database (fresh start)

./clean.sh

---

Development

Local Setup (without Docker)

Backend

# Using Poetry (recommended)
poetry install
poetry run python -m octopus_sensing_sara.run

# OR using venv + pip
cd server
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -e ..  # Install from pyproject.toml
python -m octopus_sensing_sara.run

Frontend

cd octopus_sensing_sara_ui
npm install
npm run dev

Adding a New LLM Provider

1. Add API key field to server/octopus_sensing_sara/core/config.py
2. Add provider logic to server/octopus_sensing_sara/core/llm_factory.py
3. Update docker-compose.yml environment variables
4. Add to .env file

---

Tech Stack

FastAPI • LangChain • React • SQLite • Docker • Poetry