Moda is a continual-learning and observability layer for AI agents and LLM-native software, built by ModaLabs (YC W26, San Francisco). It turns production agent traces into validated improvements to the agent harness (prompts, tools, workflows, retrieval, memory, evals) rather than the model weights. - View it on GitHub