AI Pair Programmer for Emacs

CodeTutor integrates into Emacs as a right-side panel that activates on file save. When you save, the package collects the changed diff and surrounding project context, sends it to a locally running AI backend, and surfaces a response framed as pair-programming guidance from a senior or staff engineer. The AI explains what changed, names the concept behind it, and recommends the next step. You still type every line. The panel also accepts follow-up questions via the minibuffer, maintaining recent conversation turns for continuity within a session.

The architectural boundary is the differentiating design choice. The vendor’s README is explicit that CodeTutor must not edit source files, produce patches, or generate complete ready-to-paste implementations. That constraint is not a limitation they apologize for — it is the product thesis. The feedback loop stays instructional rather than generative, which means you build the mental model, not just the file.

Because the AI backend runs locally, no code leaves your machine — a meaningful property if you are working on proprietary codebases or in an environment where routing source through a third-party API is not acceptable. Architecture notes persist to `.codetutor/ARCHITECTURE.md`, giving teams a lightweight, automatically updated record of decisions as the codebase evolves. That said, the GitHub repository shows two commits and no published release — this is described in the README itself as an early local package, which means missing documentation, rough edges in setup, and no guarantees of maintenance cadence.