Hue

Teams building AI agents need to know whether their agents reliably complete real user tasks, but testing typically happens outside the context of actual usage. Without a way to replay real requests with matching tools, data, and state, it's hard to catch failures or regressions before they reach users. Developers lack visibility into model calls, tool activity, and where coverage gaps exist in their testing. This makes it difficult to compare agent versions over time or trust that a shipped change hasn't broken something that worked before.

Hue captures real user tasks from production usage and reruns them in stateful environments with simulated APIs and MCPs that match production tools down to the edge cases. Each run gets its own isolated environment that persists every change and can be easily reset or customized. The product includes observability to trace real requests, model calls, and tool activity, eval suggestions to find tasks worth replaying and identify coverage gaps, agent replays to rerun tasks against simulated tools, and evaluations to score reruns and compare performance across agent versions and models. It integrates via an SDK or accepts OpenTelemetry traces from existing frameworks, and offers project-scoped keys with SOC 2 Type II, ISO 27001, and HIPAA compliance.