See what your agents actually did.
Instrument your application’s agent, model, and tool operations. Follow the received trace, inspect failures, and distinguish execution status from evidence of a successful outcome.
Your application. Your tools.
Observation does not depend on a particular website. Use the server-side TypeScript SDK or versioned JSON ingestion API to send operations from code you control. Traceva does not execute those tools or require a browser to embed your application.
Metadata first
The initial beta captures operation names, relationships, timing, and status—not prompts, tool arguments, outputs, or exception messages. New telemetry has a seven-day retention policy. Use safe static names: metadata can still contain sensitive information if you put it there.
What is available in this development preview?
Project-scoped server keys, cross-service trace correlation, nested evidence timelines, and filtered error, latency and usage metrics. Access is invite-only. Broader evaluation, alerting, team collaboration, and complete OpenTelemetry protocol support are not available yet.
Sign in for integration instructions · Telemetry data policy
Explore the evaluation example
The WebMCP example shows a controlled failure, diagnosis, and schema repair. It is an optional example, not a claim that arbitrary websites or tasks can already be evaluated.
Open the controlled demo