Changelog

Private Beta Launch

The initial private beta release of RAGTrace is now available to approved whitelist members. This release includes the core tracing engine, .NET SDK, and Vue.js debug panel.

Key features in this release:

• Retrieval Path Traceback with full query-to-chunk visualization
• Embedding Capture for OpenAI, Azure OpenAI, and Cohere providers
• Vector database support for Pinecone, Qdrant, and pgvector
• Real-time Vue.js debug panel with dark theme
• REST API for programmatic access to trace data

Prompt Tracing & Token Analysis

Added prompt tracing capabilities that capture the full assembled prompt template, mark context injection points, and track token distribution across sections.

• New TracePrompt() API for manual prompt instrumentation
• Token count visualization per prompt section
• Context window utilization metrics
• Fixed span timing accuracy for sub-millisecond operations

Multi-Index Comparison Engine

Introduced the A/B testing framework for comparing different RAG configurations. Run side-by-side experiments across embedding models, chunking strategies, and retrieval parameters.

• Experiment creation and management API
• Statistical significance testing for retrieval quality metrics
• Visual diff reports in the dashboard
• Improved .NET middleware performance (40% less overhead)

Vue.js Debug Panel

Released the first version of the embeddable Vue.js debug panel. Drop it into any Vue 3 application to get real-time visibility into RAG operations during development.

• Embeddable <RAGTracePanel> component
• Real-time span waterfall view
• Similarity score heatmaps
• Dark and light theme support

Performance Monitoring Dashboard

Launched the performance monitoring dashboard with real-time metrics for retrieval latency, embedding throughput, and cache hit rates.

• Latency percentile tracking (P50, P95, P99)
• Throughput charts with configurable time windows
• Alert configuration for quality regression
• Webhook integrations for Slack and Discord

Initial Alpha — Core Tracing Engine

The first internal alpha release with the core tracing infrastructure. Basic span capture, .NET middleware, and a minimal web dashboard.

• Core trace/span data model
• ASP.NET Core middleware with automatic instrumentation
• Pinecone integration (first supported vector DB)
• Basic trace list view in dashboard