cachebag/jumpy.nvim

Jumpy is a plugin for the Neovim text editor that lets a programmer ask an AI to edit code in small, reviewable chunks. The author was inspired by another tool called 99 and built jumpy because they want to keep their hands on the steering wheel even when an AI is writing code on their behalf. The repo has 14 stars and is written in Lua. The idea is to keep the human in charge of every change. Instead of letting an AI rewrite whole files or run as a separate agent, jumpy opens a prompt inside the editor, sends only the relevant lines to the AI, and shows the suggested edits as hunks inside the same buffer. The user can then accept or reject each hunk one at a time, jump to the next or previous hunk, accept everything, reject everything, or ask for a fresh attempt on the hunk under the cursor. In a comparison table, the README contrasts jumpy with other AI coding tools such as avante, sidekick, codecompanion, claude code, and aider. The pitch is that jumpy uses a per-hunk accept and reject model with no context switch out of the buffer, and that it sends only changed lines back to the model, which the author describes as more token efficient than alternatives that pass full files. Installation is done through the lazy.nvim package manager, and the plugin supports Anthropic, OpenAI, and OpenRouter as model providers via environment variables for the API key. Keybindings include leader plus j to open the prompt, bracket h to move between hunks, and leader plus a or x to accept or reject the current hunk. The author is candid that the project is early and rough. They write that they do not really know what they are doing and that the tool does not work very well yet, but they hope it will get better. The license is MIT.