Blinky is a VS Code extension focused specifically on debugging, a narrower and arguably harder problem than general code generation. Inspired by the research project SWE-agent, it triangulates the source of runtime errors by combining print-statement debugging, VS Code's Language Server Protocol, and LLM reasoning, rather than guessing at a fix from the error message alone.
The workflow is iterative: a developer describes the bug and how to reproduce it, Blinky runs a Verify step that executes the reproduction and inspects the output, and the developer reviews suggested edits before accepting or rejecting them. It needs an OpenAI API key to run and surfaces its suggestions directly in VS Code's chat interface.
It's an open-source research-adjacent project rather than a polished commercial product, with a roadmap that includes automated reproduction creation and support for additional models like LLaMA 3 and Claude, aimed at developers interested in debugging-specific AI tooling rather than general coding assistance.