Repath

Repath provides a robust system for progressive delivery of model and prompt changes. Key features include: • Real traffic splitting to new versions • Model-based quality scoring for responses • Automatic rollout advancement or rollback • Detailed logging of judgments and reasons • Seamless integration with existing SDKs like OpenAI, Anthropic, and Gemini This platform addresses the silent failure mode where model performance degrades without triggering traditional errors or alerts. It actively monitors the quality of responses from new prompt or model versions using a judge model, ensuring that any decrease in usefulness is detected and addressed immediately, long before it impacts end-users. Unlike typical dashboards that track latency and error rates, Repath focuses on the actual quality of the output. Repath integrates by simply changing one base-URL in your existing SDK. It routes a slice of live traffic to the candidate version, evaluates its performance, and makes informed decisions on whether to advance the rollout (e.g., 5% to 25% to 100%) or roll it back to 0% automatically. The entire process, including routing, recording, judging, and deciding, occurs efficiently, adding minimal overhead (~0.4 ms median). This tool is ideal for engineering and product teams who want to maintain high standards for their applications' outputs. It ensures that every change, whether to a prompt or a model, is rigorously tested against real user interactions, preventing regressions and enhancing reliability. It acts as a critical quality gate for all model-driven applications.