Install the px CLI and run px deploy to deploy the Pixie Edge Modules (PEMs) as a DaemonSet and the Kelvin aggregator; Pixie requires kernel 4.14+ and glibc 2.17+
After deployment, run px demo deploy px-sock-shop (or your own app) then open the Pixie Live UI with px live to browse built-in PxL script output
Select the px/http_data script from the UI or run px run px/http_data to see per-endpoint latency, error rates, and request/response bodies captured at the kernel level
Write a custom PxL script using px.DataFrame to query a specific table (e.g., http_events, process_stats, network_stats), filter by pod or namespace, and compute aggregations
Export Pixie data to an external observability backend by deploying the OpenTelemetry Plugin for Pixie, which converts PxL output to OTLP traces and metrics
Use px/cluster for a service map overview and px/pod to drill into CPU, memory, and network stats for a specific pod without SSH access to the node
Known gotchas
Pixie stores all data in-cluster in memory; there is no long-term retention by default — data is lost when PEM pods restart, so export critical data to an external store
PEM pods require privileged access to load eBPF programs; clusters with strict PodSecurityStandards (enforce: restricted) will reject PEM DaemonSet pods without policy exceptions
HTTP/2 and gRPC traffic inspection requires kernel 5.2+ with TLS offload support in Pixie's eBPF probes; older kernels see only TCP-level data for encrypted traffic
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp