Koredex is a solo-built autonomous debugging agent for Python developers that closes the loop between pytest failure detection and code remediation without human intervention. Developed over roughly three weeks by a single developer (GitHub: abhinav300j), the tool runs a project's full pytest suite, identifies failing tests, proposes and applies fixes autonomously, then validates each fix by checking pytest's return code — a deterministic success criterion that sidesteps the subjective self-evaluation common in many LLM coding tools. If a fix introduces new regressions, Koredex rolls it back automatically, guaranteeing the codebase is never left in a worse state than before the agent intervened.
The technical stack reflects a pragmatic indie build. A FastAPI backend handles orchestration and subprocess management, React surfaces a diff-style view of every change applied, Supabase provides lightweight persistence, and Google's Gemini API supplies the LLM reasoning that interprets test output and generates fix candidates. The choice of Gemini over OpenAI models is consistent with the cost-sensitive constraints of a solo project. The run-detect-fix-validate-rollback loop is the core agentic pattern here — and unlike tools that rely on model self-assessment, Koredex anchors every outcome to a binary pytest return code. Pass or fail. No interpretation required.
In its current form, Koredex targets a well-scoped but high-frequency slice of debugging work: dependency errors, import issues, environment misconfigurations, and simple logic bugs. The developer has openly acknowledged that LLM-based repair degrades on complex multi-file logic failures, a candid limitation that distinguishes this as an early-stage functional tool rather than a finished product. The project was shared on Hacker News as a Show HN seeking community feedback.
Koredex enters a crowded field that includes Devin, SWE-agent, and GitHub Copilot Autofix, but differentiates through its narrow pytest-specific focus and <a href="/news/2026-03-14-elastifund-open-source-ai-agent-prediction-market-trading">deterministic rollback guarantee</a>. SWE-agent, for context, targets the full SWE-bench benchmark across diverse repository types and failure classes — Koredex makes no such claim, and is more honest for it. The HN thread drew early questions about multi-file support and CI integration, which will likely shape what the developer prioritizes next.