RELEASE_NOTES_v1.3.2

ThemisDB v1.3.2 - Image Analysis AI Plugin Architecture

Release Date: 21. Dezember 2025
Focus: Image Analysis & Multi-Modal AI

---

🎉 Overview

ThemisDB v1.3.2 introduces a comprehensive image analysis AI plugin architecture that runs in parallel with LLM processing, enabling multi-modal AI workloads directly in the database. The plugin system supports multiple backend implementations with complete license compatibility.

---

🖼️ New Features

Image Analysis AI Plugin Architecture (PR #118)

Plugin System

• IImageAnalysisBackend Interface: Extensible plugin interface for image analysis backends
• ImageAnalysisManager: Centralized management for image analysis plugins
• Parallel Execution: Runs alongside LLM processing without interference
• Multi-Backend Support: Choose from multiple implementation options

Supported Backends

1. llama.cpp Vision (Primary - Recommended)

   • Native integration with llama.cpp
   • Supports vision-capable models (LLaVA, CLIP)
   • GPU acceleration support
   • License: MIT

2. ONNX Runtime

   • Microsoft's cross-platform ML inference engine
   • Wide model compatibility (CLIP, Vision Transformer)
   • CPU and GPU support
   • License: MIT

3. OpenCV DNN

   • Lightweight and battle-tested
   • Good CPU performance
   • Smaller model support
   • License: Apache 2.0

4. OpenVINO (Intel)

   • Optimized for Intel hardware
   • Excellent CPU performance
   • Wide model support
   • License: Apache 2.0

5. ncnn (Tencent)

   • Mobile-optimized
   • ARM CPU excellence
   • Minimal dependencies
   • License: BSD-3-Clause

License Compatibility

• ✅ All backends are MIT, Apache 2.0, or BSD-3-Clause
• ✅ Fully compatible with ThemisDB's MIT license
• ✅ No GPL or restrictive licenses

Example Implementation

• ONNX CLIP Plugin: Complete reference implementation included
• Demonstrates plugin interface usage
• Shows multi-backend integration patterns

Configuration

• Plugin Management: YAML-based configuration templates
• Backend Selection: Runtime backend switching
• Resource Limits: Configurable memory and GPU usage

Testing & Benchmarks

• 15+ Unit Tests: Comprehensive test coverage
• 11+ Benchmark Categories: Performance validation
  • Image encoding/decoding
  • Feature extraction
  • Similarity search
  • Multi-image batch processing
  • GPU vs CPU performance

Documentation

• 7 C++ Libraries Evaluated: Detailed comparison
• Benchmark Results: Performance metrics for each backend
• Optimization Guide: Best practices for production deployment
• Integration Examples: Complete working examples

---

🚀 Benefits

• Multi-Modal AI: Combine text (LLM) and image analysis in one database
• Flexibility: Choose the best backend for your hardware and use case
• Performance: GPU acceleration where available, optimized CPU fallback
• Extensibility: Easy to add custom backends via plugin interface
• Production Ready: Comprehensive testing and documentation

---

📚 Documentation

• Image Analysis Plugin Guide
• Backend Comparison
• ONNX CLIP Example
• Configuration Reference

---

🔄 Upgrade Notes

Build Changes

• Image analysis is optional and requires build flag:

  cmake -DTHEMIS_ENABLE_IMAGE_ANALYSIS=ON

New Dependencies (Optional)

• Choose one or more backends during build:
  • -DTHEMIS_IMAGE_BACKEND_LLAMACPP=ON (recommended)
  • -DTHEMIS_IMAGE_BACKEND_ONNX=ON
  • -DTHEMIS_IMAGE_BACKEND_OPENCV=ON
  • -DTHEMIS_IMAGE_BACKEND_OPENVINO=ON
  • -DTHEMIS_IMAGE_BACKEND_NCNN=ON

Configuration

• Add image analysis configuration to config.yaml:

  image_analysis:
    enabled: true
    backend: "llamacpp_vision"  # or "onnx", "opencv", etc.
    model_path: "/path/to/clip-model.gguf"

---

📦 Compatibility

• Backward Compatible: Yes - 100% compatible with v1.3.1
• Optional Feature: Does not affect existing functionality if not enabled
• Database Format: No changes
• API: New image analysis endpoints added (optional)
• Configuration: New optional image analysis section

---

🔗 Links

• GitHub Release
• Changelog
• PR #118