Changelog Archive: v0.3.x Series¶

[v0.3.4] - 2025-08-16¶

Hybrid Grep Command
Semantic search capability integrated into grep command
Shows both “Text Matches” (regex) and “Semantic Matches” sections
--regex-only flag to disable semantic search when needed
--semantic-limit option to control number of semantic results (default: 3)
Automatic parallel execution of regex and semantic searches for better performance
Enhanced Vector Search
Query embedding generation for accurate semantic search
Automatic embedding generator creation based on available models
Improved search accuracy with actual query vectors instead of placeholders
Default embedding support through factory pattern
Automatic Embedding Generation
Vector database auto-creation on initialization
Repair thread auto-starts when missing embeddings detected
Background embedding generation without manual intervention
Proactive database initialization to avoid “Not found” messages

Vector Database Initialization
Fixed auto-creation of vectors.db on first run
Repair thread now triggers during storage initialization
Improved embedding health detection with better sampling
Correct variable names in initialization code (metadataRepo_ vs metadataRepository_)

MCP Server
Spec-compliant stdio framing (Content-Length + CRLF) with readiness notifications
Tool schemas aligned with implementation and tests
MCP tool handlers refactored into thin adapters that map JSON ↔ service DTOs
CLI/MCP parity via shared service layer (Search, Grep, Document); MCP now delegates:
- grep_documents → GrepService
- list_documents, retrieve_document, update_document_metadata, get_by_name, cat_document, delete_by_name → DocumentService
Tracy Profiling Support
List Command Enhancements
--changes flag to show documents with recent modifications (last 24h by default)
--since option to show documents changed since specified time (ISO8601, relative, or natural formats)
--diff-tags flag for grouped change visualization showing added/modified/deleted documents
--show-deleted flag to include deleted documents in listings
File type indicators in diff-tags output showing [extension|bin/txt|category] for all files
Grep Command Enhancements
Support for searching specific file paths in addition to patterns
Improved path matching with suffix lookup for partial paths
--paths-only flag for LLM-friendly output (one file path per line)
Code Quality
Added clang-format configuration for consistent code style
Pre-commit hook for automatic code formatting
Application services layer for shared CLI/MCP functionality

MCP Server Initialization
Vector Database: Simplified embedding generation using existing repair command patterns
Stats no longer unconditionally recommends ‘yams repair –embeddings’ when embeddings already exist; guidance is shown only when count is zero
Stats Display: Improved embedding system status to show meaningful health info instead of misleading zero-activity counters
Vector Index: Fixed initialization with smart dimension detection from existing models and databases
Async Processing: Documents now get embeddings generated asynchronously in detached threads (non-blocking)
Reliability: Embedding generation now works by default, if it fails you will see yams repair --embeddings in yams status
Stats Command: Replaced confusing worker activity counters with clear system capability status
List Command Filtering: Fixed diff-tags logic to properly skip change filters when showing diff visualization
Security
Fixed command injection vulnerability in model_command.cpp by replacing system() calls with safe execvp()
Added path validation and sanitization for user inputs
Validate output paths to prevent directory traversal attacks

Preview: Vector Database (experimental)
SqliteVecBackend for persistent vector storage via the sqlite-vec extension
BLOB storage format for embeddings with transaction-safe writes across vector and metadata tables
CLI integration: enable via yams init or follow prompts shown in yams stats --health
MCP Server
Stdio transport only; HTTP framework dependencies removed

Usage
Vector search is off by default. Initialize and configure embeddings with yams init or via health prompts in yams stats --health.
Requirements
SQLite with the sqlite-vec extension available at runtime.
Limitations (preview)
Schema and APIs may change between minor versions.
Not enabled by default; indexing currently targets text documents only.
Single-process/local database only; no cross-process index sharing yet.

Critical: Missing ref_transactions table after init
Fixed incomplete schema initialization in reference counter that caused “no such table: ref_transactions” error
Added complete inline schema fallback with all required tables (ref_transactions, ref_transaction_ops, ref_statistics, ref_audit_log)
Implemented proper resource file discovery for production deployments
Added installation of SQL schema files to CMake configuration
CI/CD Workflow Simplification
Removed all cross-compilation logic from release workflow
Fixed CI build failures on Linux (gtest cache corruption) and macOS (runner detection)
Moved macOS x86_64 builds to GitHub hosted runners (macos-13)
Docker builds now use native architecture on each runner (ARM on macOS, x86 on Linux)

Build Infrastructure
All builds now use native compilation for better performance and reliability
Simplified Conan configuration by removing cross-compilation complexity
Docker workflow split into separate jobs per architecture for efficiency
Data files (magic_numbers.json, reference_schema.sql) now properly installed with binary

Apple Silicon Optimizations
New YAMS_ENABLE_APPLE_SILICON_OPTIMIZATIONS CMake option (enabled by default on ARM64 Macs)
Native -mcpu=apple-m1 optimization flag for M1/M2/M3 processors
Added -fvectorize for improved SIMD utilization on Apple Silicon

CI/CD C++ Standard Configuration
Fixed Conan profile resetting to C++17 in CI/CD workflows
Added explicit C++20 enforcement after conan profile detect --force
Fixed cross-compilation profile configurations for both CI and Release workflows