Sentry
Sentry is a platform for application error and performance monitoring. Since 2025 it offers dedicated agent tracing: a complete agent run is captured as a sequence of spans embedded in existing application traces, including cost and token breakdowns.
Sentry monitors application errors and performance and now also traces agent runs: system prompt, model calls, tool calls and handoffs appear as spans next to HTTP and database calls.
Explanation
Its particular value lies in consolidation. Many AI tools show model calls in isolation; Sentry places them next to database queries, queues and HTTP calls from the same request. That answers the question that actually arises in production: was it the model, the tool or the slow query underneath? Conversations are rendered as a readable transcript, tools reported with their own error rate and response time. Common agent frameworks are auto-instrumented. Sentry is a diagnostic tool, not an evaluation framework: it finds faults but does not systematically measure the substantive quality of an answer. For European requirements, clarify the data region upfront, and also which content ends up in error reports at all.
Marketing Relevance
For teams already using Sentry, one more tool becomes unnecessary: agent failures land where the team already looks. That matters for accountable owners because response times usually fail not on missing data but on signals scattered across several dashboards.
Example
A chat occasionally answers with an error message. The trace shows that not the model but the knowledge search tool times out in five percent of cases — the cause is a missing database index.
Common Pitfalls
Error reports can contain prompts and user input; without filtering, personal data ends up in monitoring. And agent monitoring does not replace evaluation: a run without errors can be substantively wrong.
Origin & History
Sentry began as a tool for error reporting in web applications and expanded into performance measurement and distributed tracing. As agents spread, an open beta for agent monitoring followed in June 2025, since extended with conversation views, cost breakdowns and auto-instrumentation of further frameworks.
Marketing Use Cases
Debugging tool failures and timeouts.
Overview of model cost and token usage per application.
Joined-up analysis of frontend, backend and model failures.
Alerts on rising error rates in agent runs.
Frequently Asked Questions
Do I need Sentry in addition to Langfuse?
Often yes, because the tools answer different questions: Sentry finds technical faults across the whole system, Langfuse evaluates substantive quality and cost per request. Small projects get by with one of them.
Do prompts end up in error reports?
Only if you allow it. Content should be filtered or masked before it reaches monitoring — especially customer input.
Which frameworks are captured automatically?
The common agent SDKs are auto-instrumented, including the AI SDK and OpenAI-adjacent agent libraries. For custom loops, spans can be added manually.
Related Services
Go deeper: Agentic AI Hub · Governance & compliance