Talking to Coding Agents from Neovim

In this Mitramandal.ai session, Saurabh Hirani goes under the hood of how Neovim talks to CLI‑based coding agents like Claude Code, OpenCode, Pi, and Ko, and why these patterns apply to any editor you use. He traces his own journey from Neovim to VS Code and back, driven by the realization that the best agentic UX often lives on the command line rather than in editor chat sidebars.

The talk walks through concrete integration patterns: Claude Code’s WebSocket‑based protocol using lockfiles, OpenCode’s REST + Server‑Sent Events model, the emerging Agent Client Protocol (ACP) over stdio, and a custom “prompt‑picker” plugin that lets Neovim talk to any CLI agent via tmux. Saurabh also shows how to get semantic diffs explained by an LLM, why editor–agent decoupling matters, and how lightweight, terminal‑native workflows can rival or beat heavier IDE setups.