El Yayster – a resident LLM that inhabits Emacs (github.com)

El Yayster, a new package for Emacs, introduces a unique approach by embedding a local language model (LLM) within the editor itself instead of merely treating it as an interface. This innovative setup allows El Yayster to directly perceive the user's live environment—such as open buffers and the current directory—and perform actions via gated Emacs Lisp functions. Users can interact with the model using natural language to draft and execute commands, with built-in permission prompts ensuring safety during mutating actions. This reduces dependence on cloud-based services by enabling local model operation, tested with compatibility across various OpenAI-compatible endpoints. El Yayster's significance lies in its local-first, self-discovering architecture that streamlines the interaction between the user and the AI, making it an efficient tool for developers who frequently use Emacs. The design emphasizes usability and immediacy, allowing users to issue commands without leaving the editor's context. The package supports non-blocking requests and provides observable metrics on the AI's state, including resource utilization and active sessions, enhancing user oversight. Overall, El Yayster paves the way for more integrated AI-tools within traditional text editors, presenting a playful yet powerful addition to the Emacs ecosystem.