Agent Dev Backend Api

This is an alpha release of a self-learning backend API agent that remembers what worked before. It searches past API implementations before starting new ones, pulling patterns from successful builds that scored above 85% and avoiding mistakes from failures. The GNN-enhanced context search claims 12% better accuracy for understanding dependencies between controllers and services, and it uses flash attention for processing large schemas. After each build it stores the pattern with a reward score based on test results and code quality, gradually building a reasoning bank of API implementations. It's ambitious stuff for an alpha, banking on the idea that agents get better by learning from their own history rather than starting fresh each time.