Mnemosyne - Semantic File Search

Search your files by meaning, not just filename. Drop an image into a folder, search "cherry blossoms" later, and find it instantly.

Quick Start

# 1. Setup
git clone <repo>
cd mnemosyne
uv venv
source .venv/bin/activate
uv sync

# 2. Configure (edit config.yaml)
# - Set provider to "ollama" or "gemini"
# - For Gemini: set GEMINI_API_KEY env var

# 3. Start watching
python main.py

# 4. In another terminal, add images
cp ~/Pictures/*.jpg input_images/

# 5. Search
python search.py "beautiful sunset"
python search.py --tag nature

How It Works

File Added → Watcher → AI Analysis → Embedding → Vector DB → Semantic Search

When you drop a file into input_images/:

1. Watcher detects the new file
2. AI Provider (Ollama/Gemini) analyzes and describes it
3. Embedder converts description to vector
4. LanceDB stores it for fast semantic search
5. Search finds files by meaning, not filename

Configuration

Edit config.yaml:

provider:
  type: "gemini"  # or "ollama"
  
  gemini:
    api_key: ""  # or set GEMINI_API_KEY env var
    model: "gemini-2.5-flash"
  
  ollama:
    base_url: "http://localhost:11434"
    model: "llama3.2-vision"

paths:
  input_dir: "input_images"
  storage_dir: "data"

Using Ollama (Local)

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Pull a vision model
ollama pull llama3.2-vision

# Start Ollama (keep running)
ollama serve

Using Gemini (Cloud)

# Set API key
export GEMINI_API_KEY="your-key-here"

# Or add to config.yaml

Usage

Start the Watcher

python main.py

Watches input_images/ and automatically processes new files.

Search

# Semantic search
python search.py "sunset over water"
python search.py "person smiling"

# Tag search
python search.py --tag nature
python search.py --tag portrait

# Limit results
python search.py "mountains" --limit 5

License

MIT