> ## Documentation Index
> Fetch the complete documentation index at: https://docs.hiloop.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Observability

> Run-scoped capture, platform lifecycle events, and constrained telemetry queries.

hiloop separates customer workload capture from platform runtime evidence.

## Workload capture

Wrap a host-side command with [`hiloop run`](/guides/capturing-telemetry). Depending on its options
and host support, the wrapper records network exchanges, OpenTelemetry spans/logs, and stdio. Each
event is attached to the registered run.

Every sandbox has an ambient run and a managed capture session. Explicit create commands, buffered
execs, and non-PTY SSH commands record process lifecycle and stdio; the Pod also shares a
cooperative HTTP proxy and OTLP receiver. PTY output is captured but keystrokes/input are not;
clients that bypass the proxy are not captured. The workload receives no Hiloop credential.

## Platform evidence

The control plane emits customer-safe `runtime` events for committed sandbox state and operation
transitions. Those events contain resource identity, state, failure class, and queue latency—not
command bodies, credentials, or workspace content.

Execution state and exit status remain available from the durable execution record regardless of
capture settings.

## Query

Use constrained SQL over the authenticated organization's `events` rows:

```sh theme={null}
hiloop query --run-id "$RUN_ID" --limit 100
```

The gateway allows one bounded `SELECT`, injects organization scope, and rejects unknown columns or
mutating SQL. Logical `lineage_path` fields may group related runs; they do not promise runtime
filesystem fork.

See [event model](/observability/event-model), [query telemetry](/observability/query-telemetry),
and [redaction](/observability/redaction).