All Posts

NodeLLM 1.18 adds judgments, backed by TypeSafe's Jev models. You ask typed yes/no, multiple-choice and scored questions about your data and get back calibrated probabilities instead of text. This release also makes withSchema() fully typed.

NodeLLM 1.17 closes the loop on our MCP roadmap with bidirectional sampling support, adds opt-in concurrent tool execution and additive callback stacking to core, and brings the same fine-grained tool controls to the ORM persistence layer.

NodeLLM 1.16 introduces surgical tool orchestration, AI-driven self-correction for tool failures, and the first-class support for advanced image manipulation via DALL-E and Gemini.

NodeLLM introduces native support for the Model Context Protocol (MCP), providing a standardized boundary for discovering and executing tools, resources, and prompt templates across any compliant server.

NodeLLM 1.15 introduces automated schema self-correction, enabling agents to recover from validation failures without manual intervention. This release also brings fine-grained middleware lifecycle control and hardened type safety.