Launch
pip install "plyra-guard[sidecar]"
plyra-guard serve
# → http://localhost:8765
guard.serve(host="0.0.0.0", port=8080)
The default sidecar port in config is 8080. The CLI command plyra-guard serve
uses port 8765. You can override both via the sidecar section in your YAML config.
What you can see
- Live action feed — every tool call as it happens, with verdict and latency
- Policy hit rates — which rules fire most
- Session replay — reconstruct any agent session from the audit log
- Block details — full context on every blocked call: action type, intent,
matched rule, and reason
The default CORS setting (allow_origins=["*"]) is fine for local
development. Lock it down before exposing the dashboard beyond localhost.
Query the audit log
from plyra_guard import AuditFilter, Verdict
# Get the 20 most recent blocked actions
entries = guard.get_audit_log(AuditFilter(
verdict=Verdict.BLOCK,
limit=20,
))
for entry in entries:
print(f"{entry.action_type} | {entry.verdict.value} | {entry.duration_ms}ms")
AuditFilter fields
| Field | Type | Default | Description |
|---|
agent_id | str | None | None | Filter by agent |
task_id | str | None | None | Filter by task |
verdict | Verdict | None | None | Filter by verdict type |
action_type | str | None | None | Filter by action type |
from_time | datetime | None | None | Start of time range |
to_time | datetime | None | None | End of time range |
limit | int | 100 | Max entries to return |
Explain a decision
from plyra_guard import ActionIntent
intent = ActionIntent(
action_type="file.delete",
tool_name="delete_file",
parameters={"path": "/etc/passwd"},
agent_id="default",
)
explanation = guard.explain(intent)
print(explanation)
# Human-readable description of which evaluators ran,
# which rule matched, and why the verdict was reached
Metrics snapshot
metrics = guard.get_metrics()
print(f"Total: {metrics.total_actions}")
print(f"Blocked: {metrics.blocked_actions}")
print(f"Avg risk: {metrics.avg_risk_score:.2f}")
print(f"Avg latency: {metrics.avg_duration_ms:.1f}ms")
print(metrics.to_prometheus())
