Senior
AI pair programmer for Cursor: on save, diffs the repo and uses a local LLM to explain what changed, what's at risk, and what to do next—all on-device.
Overview
A Cursor/VS Code extension that turns every save into a blast-radius check: the extension talks to a Rust daemon that diffs your repo, runs inference via Cactus (local LLM), and surfaces an Impact Panel with summary, risk level, impacted files, and suggested next actions. No data leaves your machine.
What I built
- VS Code extension (TypeScript) with save-triggered diff analysis and Impact Panel UI
- Rust daemon: NDJSON over Unix socket, diff parsing (tree-sitter ready), SQLite audit log
- Integration with Cactus for local LLM inference (on-device only)
- Explain Last Change command and configurable daemon/model paths
Highlights
- Local-first: all inference via Cactus; no code or diffs sent to the cloud
- Impact Panel: risk level, impacted files (click to open), suggested next actions