LibreChat has released version v0.8.8-rc4, introducing significant enhancements to its agentic workflows and architectural reliability. Key updates include the implementation of a Public Agents API that serves OpenAPI specifications and an interactive Swagger UI to manage inference, events, and skills. The platform now supports experimental "Attached Workspaces," allowing agents to operate within isolated environments using bounded queue waits and command timeouts to execute tasks. For enhanced observability, a new Trace Viewer enables users to inspect model conversations as sequential steps, detailing roles, tool rounds, and cost metrics. Technical improvements to Model Context Protocol (MCP) support ensure higher reliability by enabling per-request headers and coordinating OAuth refreshes across replicas. Performance optimizations have been integrated to support incremental Markdown streaming, virtualized model searching, and reduced rendering overhead for completed agent messages. Additionally, the platform expands its multimodal and generative capabilities, supporting React and Mermaid code artifacts, various image generation models like DALL-E 3 and Flux, and a robust Code Interpreter API that provides sandboxed execution for languages including Python, Node.js, Go, and Rust. These updates solidify LibreChat's position as a highly extensible, self-hostable interface for diverse AI providers ranging from OpenAI and Anthropic to local providers like Ollama and Apple MLX.