Observability - Hexel Studio

Observability gives you logs, metrics, and task events so you understand what your agents are doing in production — whether they’re failing silently, running slowly, or behaving exactly as expected.

In this section

Logs

Activity and output from your workloads.

Metrics

Performance and usage signals.

What you can observe

Signal	Tells you	Concrete example
Logs	What happened — the sequence of actions, errors, and outputs a resource produced.	An instance log shows the exact stack trace when a tool call failed.
Metrics	How fast and how loaded — latency percentiles (p50/p95/p99), throughput (req/sec), and CPU/memory against tier ceiling.	p99 latency rising from 200ms to 2s signals degradation before users report it.
Task events	Live progress — real-time state transitions as a task moves through planning and execution.	`task.submitted` → `task.state` → `task.complete` (or `task.failed`) streamed via SSE.

For real-time task visibility, see Streaming & Replay.

When to use which

Situation	Start with
”My task failed”	Task events — replay the stream to see where it stopped. Then check logs for the failing instance.
”Responses are slow”	Metrics — check latency percentiles and resource usage against tier ceiling.
”Something broke but I don’t know what”	Logs — filter by time range around the incident. Look for `request_id` in Data Platform errors.
”Am I about to hit a limit?”	Metrics — monitor throughput against rate limits and CPU/memory against tier capacity.

Access

Logs and metrics are available in the Console. Use them alongside usage to understand both behavior and cost.

Next steps

Start with Logs.

​In this section

Logs

Metrics

​What you can observe

​When to use which

​Access

​Next steps

In this section

What you can observe

When to use which

Access

Next steps