I was charged twice—refund the duplicate
Tasks from real usage
Tasks give your agent the same context and data it had during the original request.
Build agents that get the job done reliably. Turn your users’ requests into repeatable tests.
Backed by investors from
Y Combinator, Cursor, Clay, VercelProduct
Real user tasks, rerun in environments that behave like production.
Tools that match production
Simulated APIs and MCPs that match the real ones, down to the edge cases.
Stateful environments
Each run gets its own world that keeps every change. Easy to reset and customize.
Features
From real usage to measured performance.
Observability
Trace real requests, model calls, and tool activity
Eval Suggestions
Find tasks worth replaying from real usage and coverage gaps
Agent Replays
Rerun the task in a stateful environment with simulated tools
Evaluation scores over time
100%
90%
80%
70%
60%
Aug 4Aug 18Sep 1Sep 15
Evaluations
Score each rerun and compare performance across agent versions
Secure
Your agent. Your infrastructure. Your control.
- Control over traces and context
- Keys scoped per project and organization
- SOC 2 Type II, ISO 27001, and HIPAA compliant
Project settings
Project key
Hue project key, redactedCapture policy
Traces
Context
Compatibility
Keep the instrumentation you have
Traces from any OpenTelemetry tool
Hue’s SDK is one way in, and OTLP traces from any OpenTelemetry tool are another.
Language agnostic
Trace model calls and tool use with native SDKs supporting most languages.
Your existing agent stack
Spans from your framework are recognized in the convention it already uses.
Build agents that get the job done reliably.Replay a real task before you ship.
Hue