What's New in v2.3.0 ■ Enhanced Embedded Local LLM (tensai) & Performance - Upgraded the embedded local inference engine for greater stability and accuracy during model loading and text generation. - Added automatic detection and optimization for arm64 NEON SIMD environments. - Improved error reporting and diagnostics when running local models. ■ Expanded Preset Local LLM Models Added one-click downloads for a wider selection of lightweight, high-performance GGUF models: - Qwen 2.5 & Qwen 2.5 Coder (0.5B - 3B): Fast and optimized for log and code analysis - Qwen 3 (4B): Advanced reasoning model supporting step-by-step thinking - LLM-jp 3 (1.8B): Dedicated Japanese-language LLM - Gemma Series (2B / 1B / E2B): High-quality lightweight models by Google - SmolLM Series, Microsoft Phi 3.5 Mini, Llama 3.2, and more ■ Improved Model Management UI - Added an inline confirmation UI for deleting models, preventing accidental removals while maintaining a smooth workflow. - Automatically cleans up leftover model caches upon deletion to save disk space. ■ Security & Reliability Fixes - Updated underlying dependencies to resolve known vulnerabilities and enhance overall stability.

[What's New in v2.2.0] ■ Threat Detection & Compliance Audit (Sigma) Report - Built-in evaluation engine powered by industry-standard Sigma and Wazuh rules. - Automatically detects and visualizes indicators of compromise (IoC), cyber attacks, and compliance violations from your logs. - Interactive visualizations: Severity Distribution (Donut chart), MITRE ATT&CK & Compliance Tags (Bar chart), and Threat Timeline (Time-series chart). ■ Rich Out-of-the-Box Rule Packs - Pre-packaged and ready-to-use rule sets: Windows essentials/AD, Linux system/auth, Web attacks (SQLi, XSS, etc.), Network threats, Wazuh rule sets, and Compliance audit standards (PCI-DSS, NIST, GDPR, CIS). ■ External Sigma Rules Integration - Easily load and scan with your own custom Sigma rules (YAML files) or entire community rule directories. ■ AI-Powered Incident Explanation - Integrated LLM analysis provides deep contextual explanations, mapping to MITRE ATT&CK tactics, assessing compliance impact, and suggesting prioritized remediation steps. ■ Sigma CLI Tooling - Added a command-line interface (`twlogaian sigma`) for headless scanning, rule testing, datastore evaluation, and automated Wazuh-to-Sigma rule conversion. ■ General Improvements - UI refinements and minor bug fixes.

What's New in Version 2.1.0 TWLogAIAN v2.1.0 brings major enhancements, including an embedded local LLM with GPU acceleration, flexible storage engines, deep learning anomaly detection models, and powerful new analytical reports. [Embedded Local LLM & GPU Acceleration] • Built-in local LLM inference engine ("tensai") enables complete on-premises AI analysis without requiring external API keys or cloud connections. • Hardware-accelerated GPU inference support (Metal / DirectX / WebGPU). • New Model Manager interface for one-click downloading, inspecting, and managing GGUF models. • Continued full support for external LLM providers (OpenAI, Anthropic, Google Gemini, Ollama). [Advanced AI Analysis & Assistance] • AI Analysis Report: Automatically aggregates recurring error patterns and generates comprehensive incident investigations and remediation action plans. • AI Report Explanation: Generate instant analytical summaries and risk assessments directly from any report screen (Ranking, Time Series, Interval, etc.) with animated Cyber Cat feedback. • Ask AI: Conversational log investigation assistant with automatic PII masking for privacy protection. • AI Pattern Generation: Generate search regular expressions from plain language descriptions and auto-construct Grok extraction patterns from sample logs. [Flexible Storage Engines] • Choose between the Bluge full-text search engine and high-throughput columnar/key-value storage engines (Parquet, Badger, Bbolt). • When using Parquet/Badger/Bbolt, execute 100% accurate standard regular expressions directly against raw log lines. [Machine Learning & Deep Learning Anomaly Detection] • Expanded anomaly detection suite featuring AutoEncoder, LSTM neural networks, KNN, Mahalanobis Distance, and Z-Score alongside Isolation Forest and LOF. [New Analytical Reports] • Email Detailed Analysis: Visualize SPF verification (via DNS lookup), delivery delays, relay hops, and authentication metrics. • Interval Analysis: Detect periodic polling and C2 beaconing patterns with interval scatter plots and statistical indicators. • Access Delay Analysis: Identify latency bottlenecks, response spikes, and timestamp rollback anomalies. [Expanded Log Ingestion] • Added direct ingestion support for HTTP/Web APIs, Grafana Loki, Elasticsearch, and SFTP/FTP servers. [UI & Performance Improvements] • Optimized table layouts and eliminated unwanted horizontal scrollbars. • Refined settings navigation and log source management interfaces.

- Faster log importing and indexing through parallel processing optimizations. - Upgraded app security by fixing package dependency vulnerabilities. - Improved UI layouts for better responsiveness and replaced the text editor. - Re-implemented AI assist features with local LLM (ollama) support. - Reorganized documentation and added a search query guide.

Add AI Support Bug fix

Bug fix Windows log search Add TF-IDF for anomaly detection

Log Highlight

Bug fix