Release Notes
🚀 Faster, Slimmer, and More Robust
This release focuses on optimizing the core engine for speed, scale, and a seamless developer experience. Key improvements include eliminated installation bottlenecks, a refined MCP toolset, and a hardened CI pipeline for cross-platform reliability.
🌟 Key Highlights
- 10x Faster Initial Installation: The heavy embedding model download is now deferred from the initial
npm installto the interactiveopenlmlibsetup phase. This makes the initial package installation near-instant. - Intelligent Local Development: The CLI and installer now automatically detect and prioritize a local
.venvwhen running from the source repository, eliminating manual environment configuration. - Refined MCP API: All 58 MCP tools now follow a clean, consistent
verb_objectpattern (e.g.,search_findings,query_memory). Descriptions include optimized behavioral triggers to improve LLM reasoning and intent alignment. - High-Concurrency Engine: Implementation of thread-local connection pooling and bulk storage updates resolves critical N+1 disk sync bottlenecks in the memory and collaboration subsystems.
🛠️ Technical Improvements
- Performance: Memoized Python binary resolution to eliminate redundant subprocess calls during CLI startup.
- Testing: The entire test suite is now refactored to
unittest(removing thepytestdependency) to ensure 100% pass rates across Ubuntu and Windows CI environments. - Benchmarking: A new unified benchmark suite allows for continuous performance evaluation of retrieval, memory, and collaboration pipelines.
- Reliability: Hardened SQLite database operations with advanced
ESCAPEclause handling and concurrent worker thread safety.
🐞 Bug Fixes
- Fixed Claude Desktop configuration errors on non-Windows platforms.
- Resolved
ProgrammingErrorduring concurrent multi-agent tool execution. - Improved
run_ci_checks.ps1to handlepipwarnings and uninstallation edge cases gracefully.
Full Changelog: [[v0.2.0...v0.2.6]