diff --git a/CLAUDE.md b/CLAUDE.md
index 2e28dfdc..d912c9bb 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -152,7 +152,7 @@ For production deployments, migrations will be handled differently.
 - Environment config: `.env` (never overwrite without confirmation)
 - Database init: `database/init_db.sql`
 - Docker config: `docker-compose.yml` (development only)
-- Production config: Generated by `setup-opentranscribe.sh` 
+- Production config: Generated by `setup-opentranscribe.sh`
 - Frontend build: `frontend/vite.config.ts`
 
 ## AI Processing Workflow
@@ -196,7 +196,7 @@ The application now includes optional AI-powered features using Large Language M
 
 **Deployment Options:**
 - **Cloud-Only**: Use `.env` configuration with external providers (OpenAI, Claude, etc.)
-- **Local vLLM**: Run `docker compose -f docker-compose.yml -f docker-compose.vllm.yml up` 
+- **Local vLLM**: Run `docker compose -f docker-compose.yml -f docker-compose.vllm.yml up`
 - **Local Ollama**: Uncomment ollama service in `docker-compose.vllm.yml` and use same command
 - **No LLM**: Leave LLM_PROVIDER empty for transcription-only mode
 
@@ -224,8 +224,8 @@ ${MODEL_CACHE_DIR}/
 The system uses simple volume mappings to cache models to their natural locations:
 ```yaml
 volumes:
-  - ${MODEL_CACHE_DIR}/huggingface:/root/.cache/huggingface
-  - ${MODEL_CACHE_DIR}/torch:/root/.cache/torch
+  - ${MODEL_CACHE_DIR}/huggingface:/home/appuser/.cache/huggingface
+  - ${MODEL_CACHE_DIR}/torch:/home/appuser/.cache/torch
 ```
 
 ### Key Benefits
@@ -234,6 +234,36 @@ volumes:
 - **User configurable**: Simple `.env` variable controls cache location
 - **No re-downloads**: Models cached after first download (2.5GB total)
 
+## Security Features
+
+### Non-Root Container User
+
+OpenTranscribe backend containers run as a non-root user (`appuser`, UID 1000) following Docker security best practices.
+
+**Benefits:**
+- Follows principle of least privilege
+- Reduces security risk from container escape vulnerabilities
+- Compliant with security scanning tools (Trivy, Snyk, etc.)
+- Prevents host root compromise in case of container breach
+
+**Migration for Existing Deployments:**
+
+If you have an existing installation with model cache owned by root, run the permission fix script:
+
+```bash
+# Fix permissions on existing model cache
+./scripts/fix-model-permissions.sh
+```
+
+This script will change ownership of your model cache to UID:GID 1000:1000, making it accessible to the non-root container user.
+
+**Technical Details:**
+- Container user: `appuser` (UID 1000, GID 1000)
+- User groups: `appuser`, `video` (for GPU access)
+- Cache directories: `/home/appuser/.cache/huggingface`, `/home/appuser/.cache/torch`
+- Multi-stage build for minimal attack surface
+- Health checks for container orchestration
+
 ## Common Tasks
 
 ### Adding New API Endpoints
@@ -254,4 +284,4 @@ volumes:
 1. Modify `database/init_db.sql`
 2. Update SQLAlchemy models
 3. Update Pydantic schemas
-4. Reset dev environment: `./opentr.sh reset dev`
\ No newline at end of file
+4. Reset dev environment: `./opentr.sh reset dev`
diff --git a/backend/DOCKER_STRATEGY.md b/backend/DOCKER_STRATEGY.md
deleted file mode 100644
index bf201d63..00000000
--- a/backend/DOCKER_STRATEGY.md
+++ /dev/null
@@ -1,188 +0,0 @@
-# Docker Build Strategy - OpenTranscribe Backend
-
-## Overview
-
-The OpenTranscribe backend uses two Docker build strategies optimized for different use cases:
-
-1. **Dockerfile.prod** - Standard production build (currently in use)
-2. **Dockerfile.prod.optimized** - Multi-stage build for enhanced security (future use)
-
-## Current Configuration
-
-### Active Dockerfile: `Dockerfile.prod`
-
-**Base Image:** `python:3.12-slim-bookworm` (Debian 12)
-
-**Key Features:**
-- ✅ Single-stage build for faster iteration
-- ✅ CUDA 12.8 & cuDNN 9 compatibility
-- ✅ Security updates (CVE-2025-32434 fixed)
-- ✅ Root user (required for GPU access in development)
-
-**Used By:**
-- `backend` service (docker-compose.yml:80)
-- `celery-worker` service (docker-compose.yml:152)
-- `flower` service (docker-compose.yml:254)
-
-### ML/AI Stack (All cuDNN 9 Compatible)
-
-| Package | Version | Notes |
-|---------|---------|-------|
-| PyTorch | 2.8.0+cu128 | CVE-2025-32434 fixed, CUDA 12.8 |
-| CTranslate2 | ≥4.6.0 | cuDNN 9 support |
-| WhisperX | 3.7.0 | Latest with ctranslate2 4.5+ support |
-| PyAnnote Audio | ≥3.3.2 | PyTorch 2.6+ compatible |
-| NumPy | ≥1.25.2 | 2.x compatible, no CVEs |
-
-### Critical Configuration
-
-**LD_LIBRARY_PATH** (Line 28):
-```dockerfile
-ENV LD_LIBRARY_PATH=/usr/local/lib/python3.12/site-packages/nvidia/cudnn/lib:/usr/local/lib/python3.12/site-packages/nvidia/cuda_runtime/lib
-```
-
-**Why This Matters:**
-- PyAnnote diarization requires cuDNN 9 libraries
-- Libraries are in Python package directory, not system path
-- Without this, you get: `Unable to load libcudnn_cnn.so.9` → SIGABRT crash
-- Must be set at Dockerfile level (persistent, can't be overridden)
-
-## Future Strategy: Optimized Build
-
-### Dockerfile.prod.optimized (Not Yet Active)
-
-**When to Use:**
-- Production deployments requiring maximum security
-- Environments that support non-root containers
-- CI/CD pipelines with security scanning
-
-**Key Improvements:**
-
-1. **Multi-Stage Build**
-   - Stage 1 (builder): Compiles dependencies with build tools
-   - Stage 2 (runtime): Minimal image, only runtime dependencies
-   - Result: ~40% smaller image size
-
-2. **Non-Root User**
-   - Runs as `appuser` (UID 1000)
-   - Follows principle of least privilege
-   - Better for production security posture
-
-3. **Security Enhancements**
-   - No build tools in final image
-   - No curl/git (attack surface reduction)
-   - OCI-compliant labels for tracking
-   - Built-in health checks
-
-4. **Library Paths** (Adjusted for non-root)
-   ```dockerfile
-   ENV LD_LIBRARY_PATH=/home/appuser/.local/lib/python3.12/site-packages/nvidia/cudnn/lib:/home/appuser/.local/lib/python3.12/site-packages/nvidia/cuda_runtime/lib
-   ```
-
-### Migration Path
-
-**Phase 1: Current** ✅
-- Using `Dockerfile.prod` (root user)
-- Verified working with GPU/CUDA
-- All services stable
-
-**Phase 2: Testing** (Next Step)
-1. Test `Dockerfile.prod.optimized` with same workload
-2. Verify GPU access works with non-root user
-3. Confirm cuDNN libraries load correctly
-4. Run full transcription pipeline test
-
-**Phase 3: Migration**
-1. Update docker-compose.yml to use `Dockerfile.prod.optimized`
-2. Update GPU device permissions if needed
-3. Deploy to staging environment
-4. Monitor for 48 hours
-5. Production rollout
-
-## Troubleshooting
-
-### Common Issues
-
-**Problem:** `Unable to load libcudnn_cnn.so.9`
-- **Cause:** LD_LIBRARY_PATH not set
-- **Fix:** Ensure LD_LIBRARY_PATH in Dockerfile (not docker-compose)
-
-**Problem:** `Worker exited with SIGABRT`
-- **Cause:** cuDNN library version mismatch
-- **Fix:** Verify PyTorch 2.8.0+cu128 → cuDNN 9.10.2
-
-**Problem:** GPU not accessible in optimized build
-- **Cause:** Non-root user lacks GPU permissions
-- **Fix:** Add user to `video` group or use `--privileged`
-
-## Development Workflow
-
-### Local Development (with venv)
-```bash
-cd backend/
-source venv/bin/activate
-pip install -r requirements-dev.txt  # Includes testing tools
-```
-
-### Container Testing
-```bash
-# Current production build
-./opentr.sh start prod
-
-# Test optimized build (after migration)
-docker compose -f docker-compose.yml -f docker-compose.optimized.yml up
-```
-
-### Building Images
-```bash
-# Standard build
-docker compose build backend celery-worker flower
-
-# Optimized build (future)
-docker compose build -f Dockerfile.prod.optimized backend
-```
-
-## Security Considerations
-
-### Current (Dockerfile.prod)
-- ✅ Updated base image (Debian 12 Bookworm)
-- ✅ CVE-2025-32434 fixed (PyTorch 2.8.0)
-- ✅ Minimal package installation
-- ⚠️ Runs as root (required for current GPU setup)
-
-### Future (Dockerfile.prod.optimized)
-- ✅ All above, plus:
-- ✅ Non-root user execution
-- ✅ Multi-stage build (no build tools in runtime)
-- ✅ Explicit OCI labels for compliance
-- ✅ Health check integration
-
-## File Structure
-
-```
-backend/
-├── Dockerfile.prod              # Current production (in use)
-├── Dockerfile.prod.optimized    # Future optimized build
-├── requirements.txt             # Production dependencies
-├── requirements-dev.txt         # Development tools
-├── DOCKER_STRATEGY.md          # This file
-└── .dockerignore               # Excludes venv, etc.
-```
-
-## Key Takeaways
-
-1. **Always use Dockerfile.prod for now** - verified working
-2. **LD_LIBRARY_PATH is critical** - must be in Dockerfile
-3. **cuDNN 9 compatibility** - all packages updated
-4. **Optimized build is ready** - awaiting GPU permission testing
-5. **No downgrade needed** - NumPy 2.x works perfectly
-
-## Change History
-
-- **2025-10-11**: Initial strategy with cuDNN 9 migration
-  - Updated PyTorch 2.2.2 → 2.8.0+cu128
-  - Updated CTranslate2 4.4.0 → 4.6.0
-  - Updated WhisperX 3.4.3 → 3.7.0
-  - Fixed LD_LIBRARY_PATH for cuDNN libraries
-  - Removed obsolete Dockerfile.dev variants
-  - Created Dockerfile.prod.optimized for future use
diff --git a/backend/Dockerfile.prod b/backend/Dockerfile.prod
index 5766cc79..36bb99af 100644
--- a/backend/Dockerfile.prod
+++ b/backend/Dockerfile.prod
@@ -1,17 +1,24 @@
-FROM python:3.12-slim-bookworm
+# =============================================================================
+# OpenTranscribe Backend - Production Dockerfile
+# Multi-stage build optimized for security with non-root user
+# Updated with cuDNN 9 compatibility for PyTorch 2.8.0+cu128
+# =============================================================================
 
-WORKDIR /app
+# -----------------------------------------------------------------------------
+# Stage 1: Build Stage - Install Python dependencies with compilation
+# -----------------------------------------------------------------------------
+FROM python:3.12-slim-bookworm AS builder
+
+WORKDIR /build
 
-# Install system dependencies
-RUN apt-get update && apt-get install -y \
+# Install build dependencies (only in this stage)
+RUN apt-get update && apt-get install -y --no-install-recommends \
     build-essential \
-    curl \
-    ffmpeg \
-    libsndfile1 \
-    libimage-exiftool-perl \
+    gcc \
+    g++ \
     && rm -rf /var/lib/apt/lists/*
 
-# Copy requirements file
+# Copy only requirements first for better layer caching
 COPY requirements.txt .
 
 # Install Python dependencies
@@ -20,20 +27,69 @@ COPY requirements.txt .
 # CTranslate2 4.6.0+ - cuDNN 9 support
 # WhisperX 3.7.0 - latest version with ctranslate2 4.5+ compatibility
 # NumPy 2.x - fully compatible with all packages, no security issues
-RUN pip install --no-cache-dir -r requirements.txt
+# Use --user to install to /root/.local which we'll copy to final stage
+RUN pip install --user --no-cache-dir --no-warn-script-location -r requirements.txt
+
+# -----------------------------------------------------------------------------
+# Stage 2: Runtime Stage - Minimal production image with non-root user
+# -----------------------------------------------------------------------------
+FROM python:3.12-slim-bookworm
+
+# OCI annotations for container metadata and compliance
+LABEL org.opencontainers.image.title="OpenTranscribe Backend" \
+      org.opencontainers.image.description="AI-powered transcription backend with WhisperX and PyAnnote" \
+      org.opencontainers.image.vendor="OpenTranscribe" \
+      org.opencontainers.image.authors="OpenTranscribe Contributors" \
+      org.opencontainers.image.licenses="MIT" \
+      org.opencontainers.image.source="https://github.com/davidamacey/OpenTranscribe" \
+      org.opencontainers.image.documentation="https://github.com/davidamacey/OpenTranscribe/blob/master/README.md"
+
+# Install only runtime dependencies (no build tools)
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    curl \
+    ffmpeg \
+    libsndfile1 \
+    libimage-exiftool-perl \
+    libgomp1 \
+    && rm -rf /var/lib/apt/lists/* \
+    && apt-get clean
+
+# Create non-root user for security
+# Add to video group for GPU access
+RUN groupadd -r appuser && \
+    useradd -r -g appuser -G video -u 1000 -m -s /bin/bash appuser && \
+    mkdir -p /app /app/models /app/temp && \
+    chown -R appuser:appuser /app
 
+# Set working directory
+WORKDIR /app
+
+# Copy Python packages from builder stage
+COPY --from=builder --chown=appuser:appuser /root/.local /home/appuser/.local
+
+# Ensure scripts in .local are usable by adding to PATH
 # Set LD_LIBRARY_PATH for cuDNN libraries from PyTorch package
 # This ensures PyAnnote and other tools can find cuDNN 9 libraries
-# Must be set at build time to persist in the container
-ENV LD_LIBRARY_PATH=/usr/local/lib/python3.12/site-packages/nvidia/cudnn/lib:/usr/local/lib/python3.12/site-packages/nvidia/cuda_runtime/lib
-
-# Create directories for models and temporary files
-RUN mkdir -p /app/models /app/temp
+# Set cache directories to user home
+ENV PATH=/home/appuser/.local/bin:$PATH \
+    PYTHONUNBUFFERED=1 \
+    PYTHONDONTWRITEBYTECODE=1 \
+    LD_LIBRARY_PATH=/home/appuser/.local/lib/python3.12/site-packages/nvidia/cudnn/lib:/home/appuser/.local/lib/python3.12/site-packages/nvidia/cuda_runtime/lib \
+    HF_HOME=/home/appuser/.cache/huggingface \
+    TRANSFORMERS_CACHE=/home/appuser/.cache/huggingface/transformers \
+    TORCH_HOME=/home/appuser/.cache/torch
 
 # Copy application code
-COPY . .
+COPY --chown=appuser:appuser . .
+
+# Switch to non-root user
+USER appuser
+
+# Health check for container orchestration
+HEALTHCHECK --interval=30s --timeout=10s --start-period=40s --retries=3 \
+    CMD curl -f http://localhost:8080/health || exit 1
 
-# Expose port
+# Expose application port
 EXPOSE 8080
 
 # Command to run the application in production (no reload)
diff --git a/backend/Dockerfile.prod.optimized b/backend/Dockerfile.prod.optimized
deleted file mode 100644
index 3c2f9889..00000000
--- a/backend/Dockerfile.prod.optimized
+++ /dev/null
@@ -1,90 +0,0 @@
-# =============================================================================
-# OpenTranscribe Backend - Production Dockerfile (Optimized)
-# Multi-stage build optimized for security and minimal image size
-# Updated with cuDNN 9 compatibility for PyTorch 2.8.0+cu128
-# =============================================================================
-
-# -----------------------------------------------------------------------------
-# Stage 1: Build Stage - Install Python dependencies with compilation
-# -----------------------------------------------------------------------------
-FROM python:3.12-slim-bookworm AS builder
-
-WORKDIR /build
-
-# Install build dependencies (only in this stage)
-RUN apt-get update && apt-get install -y --no-install-recommends \
-    build-essential \
-    gcc \
-    g++ \
-    && rm -rf /var/lib/apt/lists/*
-
-# Copy only requirements first for better layer caching
-COPY requirements.txt .
-
-# Install Python dependencies
-# All packages now use cuDNN 9 for CUDA 12.8 compatibility
-# PyTorch 2.8.0+cu128 - includes CVE-2025-32434 security fix
-# CTranslate2 4.6.0+ - cuDNN 9 support
-# WhisperX 3.7.0 - latest version with ctranslate2 4.5+ compatibility
-# NumPy 2.x - fully compatible with all packages, no security issues
-# Use --user to install to /root/.local which we'll copy to final stage
-RUN pip install --user --no-cache-dir --no-warn-script-location -r requirements.txt
-
-# -----------------------------------------------------------------------------
-# Stage 2: Runtime Stage - Minimal production image
-# -----------------------------------------------------------------------------
-FROM python:3.12-slim-bookworm
-
-# OCI annotations for metadata
-LABEL org.opencontainers.image.title="OpenTranscribe Backend" \
-      org.opencontainers.image.description="AI-powered transcription backend with WhisperX and PyAnnote" \
-      org.opencontainers.image.vendor="OpenTranscribe" \
-      org.opencontainers.image.authors="OpenTranscribe Contributors" \
-      org.opencontainers.image.licenses="MIT" \
-      org.opencontainers.image.source="https://github.com/yourusername/transcribe-app" \
-      org.opencontainers.image.documentation="https://github.com/yourusername/transcribe-app/blob/main/README.md"
-
-# Install only runtime dependencies (no build tools, no git, no curl)
-RUN apt-get update && apt-get install -y --no-install-recommends \
-    ffmpeg \
-    libsndfile1 \
-    libimage-exiftool-perl \
-    libgomp1 \
-    && rm -rf /var/lib/apt/lists/* \
-    && apt-get clean
-
-# Create non-root user for security
-RUN groupadd -r appuser && \
-    useradd -r -g appuser -u 1000 -m -s /bin/bash appuser && \
-    mkdir -p /app /app/models /app/temp && \
-    chown -R appuser:appuser /app
-
-# Set working directory
-WORKDIR /app
-
-# Copy Python packages from builder stage
-COPY --from=builder --chown=appuser:appuser /root/.local /home/appuser/.local
-
-# Ensure scripts in .local are usable by adding to PATH
-# Set LD_LIBRARY_PATH for cuDNN libraries from PyTorch package
-# This ensures PyAnnote and other tools can find cuDNN 9 libraries
-ENV PATH=/home/appuser/.local/bin:$PATH \
-    PYTHONUNBUFFERED=1 \
-    PYTHONDONTWRITEBYTECODE=1 \
-    LD_LIBRARY_PATH=/home/appuser/.local/lib/python3.12/site-packages/nvidia/cudnn/lib:/home/appuser/.local/lib/python3.12/site-packages/nvidia/cuda_runtime/lib
-
-# Copy application code
-COPY --chown=appuser:appuser . .
-
-# Switch to non-root user
-USER appuser
-
-# Health check for container orchestration
-HEALTHCHECK --interval=30s --timeout=10s --start-period=40s --retries=3 \
-    CMD python -c "import urllib.request; urllib.request.urlopen('http://localhost:8080/health').read()" || exit 1
-
-# Expose application port
-EXPOSE 8080
-
-# Run application with auto-scaling workers (Uvicorn detects CPU cores)
-CMD ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "8080"]
diff --git a/backend/README.md b/backend/README.md
index 8e39c02a..0ba1b33d 100644
--- a/backend/README.md
+++ b/backend/README.md
@@ -1,6 +1,6 @@
 <div align="center">
   <img src="../assets/logo-banner.png" alt="OpenTranscribe Logo" width="400">
-  
+
   # Backend
 </div>
 
@@ -96,7 +96,7 @@ Required environment variables for AI processing:
 OpenTranscribe automatically caches AI models for persistence across container restarts:
 
 - **WhisperX Models**: Cached via HuggingFace Hub (~1.5GB)
-- **PyAnnote Models**: Cached via PyTorch/HuggingFace (~500MB)  
+- **PyAnnote Models**: Cached via PyTorch/HuggingFace (~500MB)
 - **Alignment Models**: Cached via PyTorch Hub (~360MB)
 - **Total Storage**: ~2.5GB for complete model cache
 
@@ -111,7 +111,7 @@ You also need to accept the user agreement for the following models:
 
 #### Troubleshooting AI Processing
 - **High GPU Memory Usage**: Try reducing `BATCH_SIZE` or changing `COMPUTE_TYPE` to `int8`
-- **Slow Processing**: Consider using a smaller model like `medium` or `small` 
+- **Slow Processing**: Consider using a smaller model like `medium` or `small`
 - **Speaker Identification Issues**: Adjust `MIN_SPEAKERS` and `MAX_SPEAKERS` if you know the approximate speaker count
 
 #### AI/ML References
@@ -152,8 +152,8 @@ backend/
 ├── scripts/                  # Utility scripts
 ├── tests/                    # Test suite
 ├── requirements.txt          # Python dependencies
-├── Dockerfile.dev           # Development container
-└── README.md               # This file
+├── Dockerfile.prod           # Production container (multi-stage, non-root)
+└── README.md                 # This file
 ```
 
 ## 🛠️ Development Guide
@@ -400,6 +400,39 @@ OPENSEARCH_URL=your-opensearch-url
 - **Search**: OpenSearch status
 - **Workers**: Celery worker health
 
+### Container Security
+
+OpenTranscribe backend follows Docker security best practices:
+
+**Non-Root User Implementation:**
+- Containers run as `appuser` (UID 1000) instead of root
+- Follows principle of least privilege for enhanced security
+- Compliant with security scanning tools (Trivy, Snyk, etc.)
+
+**Multi-Stage Build:**
+- Build dependencies isolated from runtime image
+- Minimal attack surface with only required runtime packages
+- Reduced image size and faster deployments
+
+**GPU Access:**
+- User added to `video` group for GPU device access
+- Compatible with NVIDIA Container Runtime
+- Supports CUDA 12.8 and cuDNN 9 for AI models
+
+**Model Caching:**
+- Models cached in user home directory (`/home/appuser/.cache`)
+- Persistent storage between container restarts
+- No re-downloads required after initial setup
+
+**Migration for Existing Deployments:**
+```bash
+# Fix permissions for existing model cache
+./scripts/fix-model-permissions.sh
+
+# Restart containers with new image
+docker compose restart backend celery-worker
+```
+
 ## 🤝 Contributing
 
 ### Development Process
@@ -456,4 +489,4 @@ pytest --cov=app tests/         # With coverage
 
 ---
 
-**Built with ❤️ using FastAPI, SQLAlchemy, and modern Python technologies.**
\ No newline at end of file
+**Built with ❤️ using FastAPI, SQLAlchemy, and modern Python technologies.**
diff --git a/docker-compose.prod.yml b/docker-compose.prod.yml
index 5e6b545f..aca46acb 100644
--- a/docker-compose.prod.yml
+++ b/docker-compose.prod.yml
@@ -81,8 +81,8 @@ services:
     pull_policy: always
     restart: always
     volumes:
-      - ${MODEL_CACHE_DIR:-./models}/huggingface:/root/.cache/huggingface
-      - ${MODEL_CACHE_DIR:-./models}/torch:/root/.cache/torch
+      - ${MODEL_CACHE_DIR:-./models}/huggingface:/home/appuser/.cache/huggingface
+      - ${MODEL_CACHE_DIR:-./models}/torch:/home/appuser/.cache/torch
       - backend_temp:/app/temp
     ports:
       - "${BACKEND_PORT:-5174}:8080"
@@ -163,8 +163,8 @@ services:
     restart: always
     command: celery -A app.core.celery worker --loglevel=info -Q gpu,nlp,utility,celery --concurrency=1
     volumes:
-      - ${MODEL_CACHE_DIR:-./models}/huggingface:/root/.cache/huggingface
-      - ${MODEL_CACHE_DIR:-./models}/torch:/root/.cache/torch
+      - ${MODEL_CACHE_DIR:-./models}/huggingface:/home/appuser/.cache/huggingface
+      - ${MODEL_CACHE_DIR:-./models}/torch:/home/appuser/.cache/torch
       - backend_temp:/app/temp
     environment:
       # Same environment as backend
@@ -259,8 +259,8 @@ services:
       - CELERY_BROKER_URL=redis://${REDIS_HOST:-redis}:6379/0
       - HUGGINGFACE_TOKEN=${HUGGINGFACE_TOKEN:-}
     volumes:
-      - ${MODEL_CACHE_DIR:-./models}/huggingface:/root/.cache/huggingface
-      - ${MODEL_CACHE_DIR:-./models}/torch:/root/.cache/torch
+      - ${MODEL_CACHE_DIR:-./models}/huggingface:/home/appuser/.cache/huggingface
+      - ${MODEL_CACHE_DIR:-./models}/torch:/home/appuser/.cache/torch
       - flower_data:/app
 
 volumes:
diff --git a/docker-compose.yml b/docker-compose.yml
index 7b93d40b..b96407ba 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -81,8 +81,8 @@ services:
     restart: always
     volumes:
       - ./backend:/app
-      - ${MODEL_CACHE_DIR:-./models}/huggingface:/root/.cache/huggingface
-      - ${MODEL_CACHE_DIR:-./models}/torch:/root/.cache/torch
+      - ${MODEL_CACHE_DIR:-./models}/huggingface:/home/appuser/.cache/huggingface
+      - ${MODEL_CACHE_DIR:-./models}/torch:/home/appuser/.cache/torch
     ports:
       - "5174:8080"
     healthcheck:
@@ -162,8 +162,8 @@ services:
               device_ids: ['${GPU_DEVICE_ID:-0}']
     volumes:
       - ./backend:/app
-      - ${MODEL_CACHE_DIR:-./models}/huggingface:/root/.cache/huggingface
-      - ${MODEL_CACHE_DIR:-./models}/torch:/root/.cache/torch
+      - ${MODEL_CACHE_DIR:-./models}/huggingface:/home/appuser/.cache/huggingface
+      - ${MODEL_CACHE_DIR:-./models}/torch:/home/appuser/.cache/torch
     depends_on:
       - postgres
       - redis
@@ -271,8 +271,8 @@ services:
     # No authentication required as per user requirements
     volumes:
       - ./backend:/app
-      - ${MODEL_CACHE_DIR:-./models}/huggingface:/root/.cache/huggingface
-      - ${MODEL_CACHE_DIR:-./models}/torch:/root/.cache/torch
+      - ${MODEL_CACHE_DIR:-./models}/huggingface:/home/appuser/.cache/huggingface
+      - ${MODEL_CACHE_DIR:-./models}/torch:/home/appuser/.cache/torch
 
 volumes:
   postgres_data:
diff --git a/frontend/Dockerfile.prod b/frontend/Dockerfile.prod
index fa865aef..d42bd010 100644
--- a/frontend/Dockerfile.prod
+++ b/frontend/Dockerfile.prod
@@ -31,6 +31,15 @@ RUN npm run build
 # Production stage
 FROM nginx:1.29.2-alpine3.22
 
+# OCI annotations for container metadata and compliance
+LABEL org.opencontainers.image.title="OpenTranscribe Frontend" \
+      org.opencontainers.image.description="Svelte-based Progressive Web App for AI-powered transcription" \
+      org.opencontainers.image.vendor="OpenTranscribe" \
+      org.opencontainers.image.authors="OpenTranscribe Contributors" \
+      org.opencontainers.image.licenses="MIT" \
+      org.opencontainers.image.source="https://github.com/davidamacey/OpenTranscribe" \
+      org.opencontainers.image.documentation="https://github.com/davidamacey/OpenTranscribe/blob/master/README.md"
+
 # Copy the built files from the build stage
 COPY --from=build /app/dist /usr/share/nginx/html
 
diff --git a/scripts/README.md b/scripts/README.md
index 183abcd1..17fd036c 100644
--- a/scripts/README.md
+++ b/scripts/README.md
@@ -9,6 +9,7 @@ This directory contains scripts for building Docker images, creating offline pac
 - **[install-offline-package.sh](#installation-script)** - Install OpenTranscribe on offline systems
 - **[opentr-offline.sh](#offline-management-wrapper)** - Manage offline installations
 - **[download-models.py](#model-downloader)** - Download AI models for offline packaging
+- **[fix-model-permissions.sh](#model-cache-permission-fixer)** - Fix permissions for non-root container migration
 
 ---
 
@@ -276,6 +277,93 @@ docker run --rm \
 
 ---
 
+## Model Cache Permission Fixer
+
+Script to fix model cache directory permissions when migrating to non-root container user.
+
+### Purpose
+
+OpenTranscribe backend containers now run as a non-root user (`appuser`, UID 1000) for security. Existing installations with model cache owned by root need permission updates.
+
+### When to Use
+
+Run this script if:
+- You're upgrading from a version that ran containers as root
+- Your model cache directory exists in `./models/` (or custom `MODEL_CACHE_DIR`)
+- You see permission errors when starting backend/celery containers
+
+### Prerequisites
+
+One of the following:
+- Docker installed and running
+- `sudo` access on the host system
+
+### Usage
+
+```bash
+# From project root
+./scripts/fix-model-permissions.sh
+
+# The script will:
+# 1. Read MODEL_CACHE_DIR from .env (or use default ./models)
+# 2. Check if directory exists
+# 3. Fix ownership to UID:GID 1000:1000
+# 4. Set correct permissions (755 for dirs, 644 for files)
+```
+
+### How It Works
+
+**Primary Method (Docker):**
+```bash
+docker run --rm \
+  -v ./models:/models \
+  busybox:latest \
+  chown -R 1000:1000 /models
+```
+
+**Fallback Method (sudo):**
+```bash
+sudo chown -R 1000:1000 ./models
+sudo find ./models -type d -exec chmod 755 {} \;
+sudo find ./models -type f -exec chmod 644 {} \;
+```
+
+### Output
+
+```
+OpenTranscribe Model Cache Permission Fixer
+==============================================
+
+Model cache directory: /mnt/nvm/repos/transcribe-app/models
+
+Fixing permissions using Docker container...
+✓ Permissions fixed successfully!
+
+Migration complete!
+Your model cache is now ready for the non-root container.
+```
+
+### Verification
+
+After running the script, verify permissions:
+
+```bash
+ls -la ./models/
+# Should show: drwxr-xr-x ... 1000 1000 ... huggingface
+#              drwxr-xr-x ... 1000 1000 ... torch
+```
+
+### Fresh Installations
+
+This script is **not needed** for fresh installations. The containers will automatically create the cache directories with correct ownership.
+
+### Related Documentation
+
+- [CLAUDE.md - Security Features](../CLAUDE.md#security-features) - Non-root container documentation
+- [Issue #91](https://github.com/davidamacey/transcribe-app/issues/91) - Non-root user implementation
+
+---
+
 ## Docker Build & Push Script
 
 Quick solution for building and pushing Docker images to Docker Hub locally while GitHub Actions handles automated deployments.
diff --git a/scripts/fix-model-permissions.sh b/scripts/fix-model-permissions.sh
new file mode 100755
index 00000000..319cb134
--- /dev/null
+++ b/scripts/fix-model-permissions.sh
@@ -0,0 +1,130 @@
+#!/bin/bash
+# =============================================================================
+# Fix Model Cache Permissions for Non-Root User Migration
+# =============================================================================
+# This script fixes ownership of model cache directories for the non-root
+# user implementation in OpenTranscribe backend containers.
+#
+# USAGE:
+#   ./scripts/fix-model-permissions.sh
+#
+# WHAT IT DOES:
+#   - Changes ownership of model cache directories to UID:GID 1000:1000
+#   - Ensures proper permissions (755 for directories, 644 for files)
+#   - Works with both host-mounted volumes and Docker volumes
+#
+# REQUIREMENTS:
+#   - Docker installed and running
+#   - User must have permission to run Docker commands (or use sudo)
+#
+# =============================================================================
+
+set -e  # Exit on error
+
+# Colors for output
+RED='\033[0;31m'
+GREEN='\033[0;32m'
+YELLOW='\033[1;33m'
+NC='\033[0m' # No Color
+
+# Get the script directory and project root
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
+
+echo -e "${GREEN}OpenTranscribe Model Cache Permission Fixer${NC}"
+echo "=============================================="
+echo ""
+
+# Read MODEL_CACHE_DIR from .env file if it exists
+if [ -f "$PROJECT_ROOT/.env" ]; then
+    # Source the .env file to get MODEL_CACHE_DIR
+    # Filter out comments (both full-line and inline) and empty lines
+    MODEL_CACHE_DIR=$(grep 'MODEL_CACHE_DIR' "$PROJECT_ROOT/.env" | grep -v '^#' | cut -d'#' -f1 | cut -d'=' -f2 | tr -d ' "' | head -1)
+    export MODEL_CACHE_DIR
+fi
+
+# Use default if not set
+MODEL_CACHE_DIR="${MODEL_CACHE_DIR:-$PROJECT_ROOT/models}"
+
+echo -e "${YELLOW}Model cache directory: ${MODEL_CACHE_DIR}${NC}"
+echo ""
+
+# Check if model directory exists
+if [ ! -d "$MODEL_CACHE_DIR" ]; then
+    echo -e "${YELLOW}Warning: Model cache directory does not exist yet.${NC}"
+    echo "This is normal for fresh installations. Skipping permission fix."
+    echo ""
+    exit 0
+fi
+
+# Function to fix permissions using Docker
+fix_permissions_docker() {
+    echo -e "${GREEN}Fixing permissions using Docker container...${NC}"
+
+    docker run --rm \
+        -v "$MODEL_CACHE_DIR:/models" \
+        busybox:latest \
+        sh -c "chown -R 1000:1000 /models && find /models -type d -exec chmod 755 {} \; && find /models -type f -exec chmod 644 {} \;"
+
+    if [ $? -eq 0 ]; then
+        echo -e "${GREEN}✓ Permissions fixed successfully!${NC}"
+        return 0
+    else
+        echo -e "${RED}✗ Failed to fix permissions using Docker${NC}"
+        return 1
+    fi
+}
+
+# Function to fix permissions using sudo (fallback)
+fix_permissions_sudo() {
+    echo -e "${YELLOW}Attempting to fix permissions using sudo...${NC}"
+
+    if ! command -v sudo &> /dev/null; then
+        echo -e "${RED}✗ sudo not available${NC}"
+        return 1
+    fi
+
+    sudo chown -R 1000:1000 "$MODEL_CACHE_DIR"
+    sudo find "$MODEL_CACHE_DIR" -type d -exec chmod 755 {} \;
+    sudo find "$MODEL_CACHE_DIR" -type f -exec chmod 644 {} \;
+
+    if [ $? -eq 0 ]; then
+        echo -e "${GREEN}✓ Permissions fixed successfully using sudo!${NC}"
+        return 0
+    else
+        echo -e "${RED}✗ Failed to fix permissions using sudo${NC}"
+        return 1
+    fi
+}
+
+# Try Docker method first
+if command -v docker &> /dev/null; then
+    if fix_permissions_docker; then
+        echo ""
+        echo -e "${GREEN}Migration complete!${NC}"
+        echo "Your model cache is now ready for the non-root container."
+        exit 0
+    fi
+fi
+
+# Fallback to sudo if Docker failed
+echo ""
+echo -e "${YELLOW}Docker method failed, trying sudo...${NC}"
+if fix_permissions_sudo; then
+    echo ""
+    echo -e "${GREEN}Migration complete!${NC}"
+    echo "Your model cache is now ready for the non-root container."
+    exit 0
+fi
+
+# If both methods failed
+echo ""
+echo -e "${RED}Failed to fix permissions!${NC}"
+echo ""
+echo "Manual steps:"
+echo "1. Run the following command:"
+echo "   sudo chown -R 1000:1000 $MODEL_CACHE_DIR"
+echo "2. Or use Docker:"
+echo "   docker run --rm -v $MODEL_CACHE_DIR:/models busybox chown -R 1000:1000 /models"
+echo ""
+exit 1