LangChain's Deep Agents framework now integrates with Mainbrella sandbox adapter, enabling AI agents to safely execute code and manage files in Linux containers. This integration allows agents to inspect inputs, write scripts, execute them, handle errors, and produce outputs without managing infrastructure manually. The adapter follows existing extension patterns and enables multi-agent workflows with shared workspaces.
rag-debugger v0.3.0 is a debugging tool for RAG retrieval pipelines that identifies missing knowledge, wrong chunk selections, and retrieval gaps through query decomposition, reranking, and a local dashboard. It integrates with LangChain, LlamaIndex, and custom retrievers via simple Python decorators and wrapping functions, using Google Gemini as the default LLM backend.