Add a local AI code refactor agent with memory

This project proposes a lightweight local AI code refactor agent that learns from previous refactoring attempts.

The goal is to explore experience-driven memory in code refactoring: storing failed outcomes, extracting reusable constraints, and adapting the refactor strategy based on what has previously gone wrong. This is a small, self-contained demo app that fits the repo’s pattern of practical AI app examples.

Proposed scope Local Streamlit app Code refactor workflow driven by an LLM Session-based memory for outcomes and learned constraints Strategy adaptation after repeated failures Minimal test coverage for memory and outcome tracking Why this matters The repo currently contains many conversation-memory apps, but this adds a more task-oriented memory pattern: learning from execution failures and reusing those lessons during later refactors.

Expected outcome A working local demo that:

accepts a code snippet and refactor task stores outcomes and constraints retrieves relevant past lessons improves subsequent refactor attempts using memory Notes This is intentionally a demo/tutorial project It should remain local-first and avoid unsupported claims about production-grade persistence or benchmarked accuracy This is a strong issue because it is clear, scoped, and aligned with the repo’s current direction.