screenshot-memory

Search screenshots by what's in them.

All your screenshots.

Screenshot 2024-01-15 at 10.32.41.png
Screenshot 2024-01-15 at 10.33.02.png
Screenshot 2024-01-14 at 09.15.33.png
...4,844 more

Now you can.

$ ssm find "that stripe error"

Found 1 result · 34ms

1. Screenshot 2024-01-15 at 10.32.41.png · 89%
   "Error: Your card was declined. Insufficient funds."
   2 weeks ago

What is this

A CLI that searches your screenshots by what's in them. Text, images, anything.

• OCR extracts text from screenshots
• AI vision describes photos (local, runs on your machine)
• Semantic search finds things by meaning, not just keywords using memvid

No cloud. No API keys. Everything runs locally.

Install

git clone https://github.com/memvid/screenshot-memory.git
cd screenshot-memory
./setup.sh

The setup script handles everything - Bun, Ollama, the AI model, all of it.

To uninstall:

./setup.sh --remove

Try it out

We included sample screenshots to test with:

# Index the examples
ssm index ./examples

# Search for stuff
ssm find "connection error"
ssm find "stripe"
ssm find "404"

Usage

# Index your screenshots once
ssm index ~/Screenshots

# Find stuff
ssm find "kubernetes error"
ssm find "that figma mockup"
ssm find "slack conversation with john"

# Photos work too
ssm find "red car"
ssm find "sunset at the beach"

# Watch for new screenshots
ssm watch

The photo thing

If you want to search photos by what's in them (not just text), you need Ollama running locally:

brew install ollama
ollama pull llava-phi3
ollama serve

First search takes ~6 seconds (model loading). After that it's ~1 second.

Commands

Command	What it does
ssm index <dir>	Index screenshots from a directory
ssm find <query>	Search for screenshots
ssm watch	Auto-index new screenshots
ssm stats	Show index info
ssm ocr <image>	Test OCR on an image
ssm caption <image>	Test AI description on an image

Options

ssm index ~/Screenshots --force    # Re-index everything
ssm index ~/Screenshots -w 8       # More OCR workers = faster

ssm find "error" --mode lex        # Exact text match
ssm find "error" --mode sem        # Semantic/meaning match
ssm find "error" -k 10             # More results
ssm find "error" --open            # Open first result
ssm find "error" --json            # JSON output

How fast

What	Time
Index 1 screenshot	~2s
Index 1 photo (with AI)	~1s (warm) / ~6s (cold)
Search	<100ms

"Cold" = first run after starting Ollama. Keep it warm with OLLAMA_KEEP_ALIVE=3600.

How it works

Screenshot → OCR → Text ────────────┐
                                    ├→ Embeddings → Index
Photo → AI Vision → Description ────┘
                                          ↓
              Query → Search → Results

All stored in a single .mv2 file. Powered by memvid.

Config

Lives at ~/.config/screenshot-memory/config.json:

{
  "directories": ["~/Screenshots"],
  "ocr": { "workers": 4 },
  "search": { "defaultMode": "auto", "defaultLimit": 5 }
}

Troubleshooting

"No screenshots indexed" → Run ssm index ~/Screenshots first

OCR is slow → Use more workers: ssm index -w 8

No image preview → brew install chafa

Photos not being described → Make sure Ollama is running: ollama serve

License

MIT

---

Built on memvid