Respan

Respan is an LLM observability and evaluation platform designed to help teams route, trace, monitor, and improve AI agent workflows. It provides a unified gateway through which all LLM calls can be directed, supporting over 500 models via a single API. Every request is logged and visualized in detailed trace trees that show latency per span, making it possible to identify bottlenecks and errors at a glance. The platform also offers dashboards for tracking usage, cost, tokens, and error rates, with the ability to slice metrics by model, API key, or user.

Beyond observability, Respan includes built-in evaluation capabilities that allow teams to combine human review, automated code checks, and LLM judges in customizable workflows. This enables continuous assessment of output quality against metrics like faithfulness, and can be applied to sampled production traffic. The platform also supports advanced traffic management features such as automatic fallback if a model errors or rate-limits, load balancing across API keys, retry with backoff, and caching of repeat prompts to reduce latency and cost.

Respan is positioned as an end-to-end LLM engineering platform that connects findings from monitoring and evaluation to concrete improvements. It includes version control and rollout logic for prompts and models, allowing teams to promote changes from the UI to production with confidence. The platform appears to be designed for teams already deploying or developing LLM-based products who need more visibility and control over their AI infrastructure.