Your closest moments deserve better than a forgotten folder. Rediscover and relive your entire photo library with natural language search, face recognition & story timelines — powered by local Vision AI. No cloud, no telemetry, no compromise.
-orange.svg){kind=icon}
-purple.svg){kind=icon}
-red.svg){kind=icon}
- 🔒 100% Private & Strictly Offline — Zero cloud dependencies, zero external telemetry. All neural inference, facial detection, vector embeddings, and reverse geocoding run entirely on your local machine.
- 🖥️ Native Desktop Application (Tauri v2) — High-performance native desktop shell (
PixelMemory.exe/ macOS App) featuring a frameless glass header, custom window controls, and native File Explorer/Finder directory pickers. - 🧠 Deep Visual Understanding (Moondream2) — Pre-configured to use Moondream2 (
moondream) via Ollama for ultra-fast GPU visual description (~0.8s/photo) capturing scene categories, objects, actions, clothing, colors, and mood. - 🗣️ Smart Natural Language Query Understanding — Intelligent query intent parser automatically decomposes queries like "photos in Tokyo last summer with Alice" into semantic visual vectors, location filters, date/time ranges, and recognized people.
- 🗺️ Interactive Geographic Map View — Visualizes your photos on an interactive dark-mode world map with geographic clustering based on EXIF GPS metadata.
- 📖 Personal Story Timelines (Gemma 4 2B) — Pre-configured to use Gemma 4 (
gemma4:e2b) to transform chronological photo groups and memories into warm, personal first-person journal narratives. - 👤 On-Device Face Recognition & Clustering — Built-in YuNet face detection and SFace deep facial embeddings for tagging friends, family, and pets without cloud biometric databases.
- 📍 Offline Reverse Geocoding — Automatically extracts EXIF GPS coordinates and maps them to human-readable place names (city, region, country) with zero network calls.
- 🔍 Visual Similarity / "More Like This" — Instant vector nearest-neighbor search to find visually and contextually similar memories from any photo in your archive.
- 📷 Deep EXIF & Camera Metadata Inspector — Full metadata breakdown showing camera model, lens, focal length, aperture, shutter speed, ISO, timestamp, and AI caption tags.
- ⚡ Incremental Ingest & Folder Sync — High-speed sequential photo indexing with file hash change detection, deleted photo pruning, and live ETA progress dock.
PixelMemory is out-of-the-box optimized for consumer GPUs, Apple Silicon, and modern CPUs:
| Capability | Model / Engine | Speed / Resource | Purpose |
|---|---|---|---|
| Vision (VLM) | moondream (Moondream2 1.8B) |
~0.8s/photo · 1.8 GB VRAM | Comprehensive visual captioning for semantic search |
| Stories (LLM) | gemma4:e2b (Gemma 4 2B) |
~1-2s · ~2 GB VRAM | Warm first-person daily travel journals and narratives |
| Query Intent | Fast NLP & Intent Parser | Instant (<1ms) · Local | Automatic extraction of dates, places, people, and semantics |
| Embeddings | all-MiniLM-L6-v2 |
~15ms · CPU / GPU | 384-dimensional dense semantic vector space (Zvec) |
| Vector Engine | Zvec Index | Instant (<5ms) · RAM/Disk | Fast cosine similarity vector search and retrieval |
| Face Detection | YuNet + SFace (ONNX) | Real-time · CPU / GPU | 128-dimensional cosine face clustering |
| Geocoding | Reverse Geocoder | Instant (<1ms) · Local DB | Offline GPS to City/Country mapping |
| Desktop Shell | Tauri v2 (Rust) | Native binary (<15MB) | Lightweight native window with glass styling & OS bridges |
PixelMemory uses Ollama for local GPU acceleration of Moondream2 and Gemma 4. Follow the setup steps below for your operating system:
-
Install Ollama:
- Option A (One-command with winget):
winget install -e --id Ollama.Ollama
- Option B (Installer): Download and run the installer from ollama.com/download/windows.
- Option A (One-command with winget):
-
Start Ollama: Launch Ollama from your Start Menu. A llama icon will appear in your Windows System Tray (near the clock).
-
Pull the Default Models: Open PowerShell or Command Prompt and run:
ollama pull moondream ollama pull gemma4:e2b
-
Install Ollama:
- Option A (Homebrew):
brew install --cask ollama
- Option B (Direct Download):
Download the
.zipfrom ollama.com/download/mac and dragOllama.appinto/Applications.
- Option A (Homebrew):
-
Start Ollama: Launch Ollama from Applications or Spotlight. An Ollama menu bar icon will appear at the top of your screen.
-
Pull the Default Models: Open Terminal and run:
ollama pull moondream ollama pull gemma4:e2b
To confirm both models are ready, run:
ollama listYou should see moondream:latest and gemma4:e2b listed.
- Python 3.10–3.12 (3.13+ is not yet supported by all dependencies such as
torchandzvec) - Ollama installed and running (see Ollama Setup Guide above)
- macOS only: Xcode Command Line Tools are required to compile native Python packages:
xcode-select --install
Note: If
pillow-heiffails to install (needed for HEIC/HEIF photo support), install the system library:brew install libheif
- Linux only: OpenCV requires system libraries on headless setups:
sudo apt install -y libgl1-mesa-glx libglib2.0-0
PixelMemory provides automated setup scripts that configure the Python virtual environment and check your tools:
-
Clone the repository:
git clone https://github.com/xklabs-AI/PixelMemory.git cd PixelMemory
-
Run the automated setup: Double-click
setup.bator run in PowerShell:.\setup.bat(Or using PowerShell:
.\setup.ps1)Zero-Friction Tip: If you simply run
.\launch.baton a fresh system, it will automatically detect that setup is needed and configure.venvfor you!
-
Clone the repository:
git clone https://github.com/xklabs-AI/PixelMemory.git cd PixelMemory -
Run the automated setup:
chmod +x setup.sh launch.sh ./setup.sh
PixelMemory can run either as a Native Desktop Application or as a Local Web Platform.
The desktop mode provides native OS window dragging, glass styling, and native file dialogs:
- Windows:
.\launch.bat --desktop # or: .\launch.ps1 -desktop
- macOS / Linux:
./launch.sh --desktop
Note
Desktop Mode Prerequisites: Desktop mode requires Rust/Cargo and Node.js 18+. Install Rust via:
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
source ~/.cargo/envOn Windows, Microsoft C++ Build Tools are also required. Then install frontend dependencies:
npm installIf Rust is not present, PixelMemory will inform you and gracefully offer to run in Web Browser Mode.
Warning
macOS 27+ SDK Compatibility: The macOS 27.0 SDK introduces new architecture identifiers (arm64e.x1) that the current stable Rust toolchain does not yet recognize. If the desktop build fails with linker errors referencing unknown architecture, use Web Browser Mode (./launch.sh) as a fully functional alternative until Rust ships an updated toolchain.
Runs the high-speed local server and automatically opens your default web browser:
- Windows:
.\launch.bat - macOS / Linux:
./launch.sh
Default URL: 👉 http://localhost:8642
Want to try PixelMemory instantly without waiting for your photo library to index?
- Windows:
.\launch.bat --demo
- macOS / Linux:
./launch.sh --demo
This seeds 12 curated memories (landscapes, birthdays, pets, food, travel) with GPS and embeddings in ~2 seconds so you can test natural language search right away!
To verify your hardware compute, Ollama status, and database health at any time:
# Check system status
python launch.py --status
# Interactive setup and model doctor
python launch.py --doctorYou can import photos directly through the UI:
- Open PixelMemory (Desktop App or Browser).
- Click 📁 Ingest Photos or click Browse Folder... (which opens your native OS folder chooser).
- Select your photo directory (e.g.,
D:\Photosor/Users/name/Pictures). - Click Start Ingestion. The floating progress dock will monitor progress in the background while you continue searching!
Settings can be customized in backend/config.py:
| Setting | Default | Description |
|---|---|---|
DEFAULT_VLM_MODEL |
"moondream" |
Primary Vision model for image captioning |
STORY_LLM_MODEL |
"gemma4:e2b" |
Primary LLM for travel story generation |
OLLAMA_HOST |
"http://localhost:11434" |
Local Ollama API server endpoint |
DATA_DIR |
~/.pixelmemory |
Library database, thumbnails, and Zvec vector store (auto-created on first run; grows with library size) |
PORT |
8642 |
Local backend port |
HOST |
"0.0.0.0" |
Network bind address |
- 💬 Chat with a Photo — Ask questions about any photo in your library and get natural, conversational answers
- 👨👩👧👦 Chat with a Family Member — Select a tagged person and discover what they love, where they've been, and who they spend time with — all from your photo history
- 🔍 Duplicate Photo Detection — Find and manage duplicate or near-duplicate photos across your library
- 📦 Standalone Windows Installer — One-click setup, no Python or Rust required
- 🎞️ Slideshow Generator — Turn a selection of photos into an animated slideshow with music
- Zero Cloud: 100% of image bytes, facial recognition embeddings, and EXIF coordinates remain strictly on your local disk.
- No Telemetry: No tracking cookies, analytics pings, or cloud API calls.
- Gitignored Libraries: Your photos, database, and thumbnails are excluded from version control by default.
PixelMemory is open-source software licensed under the Apache 2.0 License.