diff --git a/CLAUDE.md b/CLAUDE.md
index ef053a8..6c05015 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -25,6 +25,7 @@ Root `compose.yaml` uses Docker Compose's `include` directive to orchestrate mul
 - **kit**: Unified toolkit with Vert file converter and miniPaint image editor (path-routed)
 - **jelly**: Jellyfin media server with hardware transcoding
 - **drop**: PairDrop peer-to-peer file sharing
+- **ai**: AI infrastructure with Open WebUI, Crawl4AI, and pgvector (PostgreSQL)
 - **restic**: Backrest backup system with restic backend
 - **netdata**: Real-time infrastructure monitoring
 - **sablier**: Dynamic scaling plugin for Traefik
@@ -429,6 +430,64 @@ PairDrop peer-to-peer file sharing service:
 
 **Note**: PairDrop is stateless and doesn't require backups as no data is persisted. All transfers happen directly between devices.
 
+### AI Stack (ai/compose.yaml)
+AI infrastructure with Open WebUI, Crawl4AI, and dedicated PostgreSQL with pgvector:
+- **ai_postgres**: PostgreSQL 16 with pgvector extension exposed internally
+  - Dedicated database instance for AI/RAG workloads
+  - pgvector extension for vector similarity search
+  - scram-sha-256 authentication
+  - Health check monitoring
+  - Data persisted in `ai_postgres_data` volume
+
+- **webui**: Open WebUI exposed at `ai.pivoine.art:8080`
+  - ChatGPT-like interface for AI models
+  - Claude API integration via Anthropic API (OpenAI-compatible endpoint)
+  - PostgreSQL backend with vector storage (pgvector)
+  - RAG (Retrieval-Augmented Generation) support with document upload
+  - Web search capability for enhanced responses
+  - SMTP email configuration via IONOS
+  - User signup enabled
+  - Data persisted in `ai_webui_data` volume
+
+- **crawl4ai**: Crawl4AI web scraping service (internal API, no public access)
+  - Optimized web scraper for LLM content preparation
+  - Internal API on port 11235 (not exposed via Traefik)
+  - Designed for integration with Open WebUI and n8n workflows
+  - Data persisted in `ai_crawl4ai_data` volume
+
+**Configuration**:
+- **Claude Integration**: Uses Anthropic API with OpenAI-compatible endpoint
+- **API Base URL**: `https://api.anthropic.com/v1`
+- **RAG Embedding**: OpenAI `text-embedding-3-small` model
+- **Vector Database**: pgvector for semantic search
+- **Web UI Name**: Pivoine AI
+
+**Database Configuration**:
+- **User**: `ai`
+- **Database**: `openwebui`
+- **Connection**: `postgresql://ai:password@ai_postgres:5432/openwebui`
+
+**Usage**:
+1. Access https://ai.pivoine.art to create an account
+2. Configure Claude API key in settings (already configured server-side)
+3. Upload documents for RAG-enhanced conversations
+4. Use web search feature for current information
+5. Integrate with n8n workflows for automation
+
+**Integration Points**:
+- **n8n**: Workflow automation with AI tasks (scraping, RAG ingestion, webhooks)
+- **Mattermost**: Can send AI-generated notifications via webhooks
+- **Crawl4AI**: Internal API for advanced web scraping
+- **Claude API**: Primary LLM provider via Anthropic
+
+**Future Enhancements**:
+- GPU server integration (IONOS A10 planned)
+- Additional AI models (Whisper, Stable Diffusion)
+- Enhanced RAG pipelines with specialized embeddings
+- Custom AI agents for specific tasks
+
+**Note**: All AI volumes are backed up daily at 3 AM via Restic with 7 daily, 4 weekly, 6 monthly, and 2 yearly retention.
+
 ### Netdata (netdata/compose.yaml)
 Real-time infrastructure monitoring and alerting:
 - **netdata**: Netdata monitoring agent exposed at `netdata.pivoine.art:19999`
@@ -471,7 +530,7 @@ Backrest backup system with restic backend:
   - Prune: Weekly (Sundays at 2 AM) - removes old snapshots per retention policy
   - Check: Weekly (Sundays at 3 AM) - verifies repository integrity
 
-**Backup Plans** (11 automated daily backups):
+**Backup Plans** (16 automated daily backups):
 1. **postgres-backup** (2 AM daily)
    - Path: `/volumes/core_postgres_data`
    - Retention: 7 daily, 4 weekly, 6 monthly, 2 yearly
@@ -528,6 +587,14 @@ Backrest backup system with restic backend:
     - Path: `/volumes/jelly_config`
     - Retention: 7 daily, 4 weekly, 6 monthly, 2 yearly
 
+15. **netdata-backup** (10 AM daily)
+    - Path: `/volumes/netdata_config`
+    - Retention: 7 daily, 4 weekly, 3 monthly
+
+16. **ai-backup** (3 AM daily)
+    - Paths: `/volumes/ai_postgres_data`, `/volumes/ai_webui_data`, `/volumes/ai_crawl4ai_data`
+    - Retention: 7 daily, 4 weekly, 6 monthly, 2 yearly
+
 **Volume Mounting**:
 All Docker volumes are mounted read-only to `/volumes/` with prefixed names (e.g., `backup_core_postgres_data`) to avoid naming conflicts with other compose stacks.
 
@@ -553,6 +620,9 @@ Key variables defined in `arty.yml` and overridden in `.env`:
 - `TRACK_APP_SECRET`: Umami analytics secret
 - `MATTERMOST_WEBHOOK_URL`: Incoming webhook URL for infrastructure notifications (stored in `.env` only)
 - `WATCHTOWER_NOTIFICATION_URL`: Shoutrrr format URL for container update notifications
+- `AI_DB_PASSWORD`: AI PostgreSQL database password
+- `AI_WEBUI_SECRET_KEY`: Open WebUI secret key for session encryption
+- `ANTHROPIC_API_KEY`: Claude API key for AI functionality
 
 ## Volume Management
 
@@ -568,6 +638,8 @@ Each service uses named volumes prefixed with project name:
 - `vault_data`: Vaultwarden password vault (SQLite database)
 - `joplin_data`: Joplin note-taking data
 - `jelly_config`: Jellyfin media server configuration
+- `ai_postgres_data`, `ai_webui_data`, `ai_crawl4ai_data`: AI stack databases and application data
+- `netdata_config`: Netdata monitoring configuration
 - `restic_data`, `restic_config`, `restic_cache`, `restic_tmp`: Backrest backup system
 - `proxy_letsencrypt_data`: SSL certificates