Claude Self-Reflect v3.3.0 Release - Major Architecture Overhaul

[ramakay]

Claude Self-Reflect v3.3.0

Updated: September 12, 2025 | Version: 3.3.0

Executive Summary

Claud Self-Reflect v3.3.0 represents a significant architectural milestone with a complete server modularization that reduces codebase complexity by 68% while introducing critical performance fixes and new temporal analysis capabilities. This release addresses key stability issues and enhances the semantic memory system with production-ready infrastructure.

Key Features

Server Architecture Overhaul

• Complete modularization from 2,321 to 728 lines (68% reduction)
• Eliminated circular references causing 100% CPU usage
• Improved maintainability and testing coverage
• Enhanced error handling and logging systems

Critical Bug Fixes

• Fixed store_reflection dimension mismatch between local FastEmbed (384d) and Voyage AI (1024d) embeddings
• Resolved 100% CPU usage caused by circular import dependencies
• Corrected search functionality across all 15+ MCP tools
• Fixed real-time indexing performance bottlenecks

New Temporal Tools Suite

• get_recent_work: Retrieve conversations from specified time periods
• search_by_recency: Time-weighted semantic search with decay factors
• get_timeline: Chronological conversation analysis and insights

Real-Time Indexing

• Smart interval processing: 2-second intervals for active files, 60-second for standard
• Automatic conversation detection and vector indexing
• Support for both local FastEmbed and cloud Voyage AI embeddings

Production Infrastructure

• Precompact hook functionality for vector database optimization
• Enhanced monitoring and health check capabilities
• Improved Docker container orchestration
• Comprehensive logging and debugging tools

Technical Specifications

Embedding Models

• Local: FastEmbed all-MiniLM-L6-v2 (384 dimensions)
• Cloud: Voyage AI voyage-3 (1024 dimensions)
• Automatic detection and compatibility handling

Vector Database

• Qdrant vector storage with per-project collections
• Cross-collection semantic search capabilities
• Time-based decay scoring (90-day half-life)

MCP Server

• Python-based using FastMCP framework
• 15+ operational tools for semantic search and reflection
• Support for both user-scoped and project-scoped configurations

Performance Metrics

• Search latency: <100ms for typical queries
• Memory usage: Optimized for production workloads
• Import throughput: Streaming support for large conversation datasets

Breaking Changes

Embedding Dimension Handling

• store_reflection now automatically detects embedding type
• Existing reflections remain compatible
• No manual migration required

MCP Tool Updates

• All tools now require proper embedding type detection
• Enhanced error messages for troubleshooting
• Improved parameter validation

Installation

Quick Start

npm install -g claude-self-reflect@3.3.0

Manual Installation

git clone https://github.com/ramakay/claude-self-reflect.git
cd claude-self-reflect
npm run setup-wizard

Docker Deployment

docker compose up -d

Documentation

• Setup Guide: README.md
• MCP Reference: MCP_REFERENCE.md
• Architecture Documentation: docs/architecture/
• Troubleshooting: docs/troubleshooting/

Support

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Documentation: Project Documentation

For technical questions, feature requests, or bug reports, please use our GitHub Issues tracker. The community actively monitors and responds to all inquiries.