feat: Implement knowledge import feature for characters

- Added KnowledgeImport page for importing character knowledge from URLs.
- Integrated URL validation and error handling for unsupported websites.
- Created API endpoints for importing content from URLs and retrieving character knowledge.
- Enhanced VectorStoreService with logging and error handling for vector memory storage.
- Updated frontend to display knowledge sources and manage them effectively.
- Added support for fetching recent character knowledge as a fallback in similarity searches.
- Updated OpenAPI documentation to reflect new import functionality.

README.md

@@ -186,6 +186,45 @@ docker-compose exec backend pnpm db:migrate
 
 **Note:** An external reverse proxy (nginx, Traefik, etc.) is expected for SSL termination and routing. See [deployment.md](doc/deployment.md) for configuration examples.
 
+## Troubleshooting
+
+### Memory Issues (JavaScript heap out of memory)
+
+The local embedding model uses TensorFlow.js which can consume significant memory. If you encounter OOM errors:
+
+1. **Increase Node.js memory limit:**
+   ```bash
+   # In apps/backend/.env
+   NODE_OPTIONS=--max-old-space-size=768
+   
+   # Or when starting manually
+   node --max-old-space-size=768 dist/main
+   ```
+
+2. **Use low-memory mode (for 512MB RAM VPS):**
+   ```bash
+   cd apps/backend
+   pnpm run start:low-memory
+   ```
+
+3. **Use quantized model (4x smaller memory):**
+   ```bash
+   # In apps/backend/.env
+   EMBEDDING_QUANTIZED=true
+   ```
+
+4. **Limit import size:** The system automatically:
+   - Limits content to ~30KB per import
+   - Limits to 30 chunks maximum
+   - Processes chunks sequentially with delays
+   - Adds GC pauses between chunks
+
+### System Requirements
+
+- **Minimum:** 512MB RAM (with low-memory configuration)
+- **Recommended:** 1GB+ RAM for comfortable operation
+- **Import processing:** Large imports may take longer on low-memory systems
+
 ## License
 
 MIT