Local-only observability stack
Observability
See the traces without sending them anywhere
A loopback-bound OpenTelemetry pipeline for the Tier 0 family: OTLP traces land in a collector and are stored on local disk by Tempo. No external dial-out.
See how Observability connects to Agent Ops and the rest of the family.
Problem
The data boundary
Agent runs produce useful traces, but sending them to a hosted APM service creates a data boundary and an operating cost the local stack does not need.
Approach
Keep the pipeline local
The collector receives OTLP, Tempo stores traces on disk, and every service binds to loopback. The retention boundary is the machine, not a vendor account.
What it does
01
OTLP native
The collector speaks OTLP directly; no proprietary agent SDK.
02
Local disk
Tempo persists traces locally, so the retention boundary is the machine.
03
Loopback only
No external dial-out; the pipeline is not a hosted service.
04
Citable config
Configuration is ready and can be reviewed as code; container startup is pending registry recovery.
What is real today
Facts taken from the current catalog description.
- Protocol
- OTLP
- Backend
- Tempo
- Binding
- loopback
- State
- config ready
Boundary
It is not a hosted APM product. It is the local observability layer for the Tier 0 family; container startup is still pending registry recovery.
Where it sits
→ This product
feeds · ProposedAgent Ops
Run metrics
Evidence (1)
agent-ops/docs/prd/Agent-Ops-多Agent角色与团队-PRD-2026-11-19.mdPRD M3 future scope: run metrics exported to Observability.
See how Observability connects to Agent Ops and the rest of the family.
Open the product map