Overwatcher.dev — Autonomous Fleet & Service Health
An overwatcher is the layer that notices before a human does: systemd restarts, SLO burn, health endpoints, resource ceilings. This page specifies the overwatcher contract and ships a live policy checker that scores a unit file against it.
systemd Unit + SLO Policy Checker
Paste a systemd unit (or drop-in) and SLO knobs. The checker validates restart policy, health probing, resource limits and error-budget burn — the same rules an autonomous overwatcher applies before paging a human.
Fill the fields and press Check Policy.
The overwatcher contract
- Self-heal first — Restart=always with RestartSec backoff; on-failure is acceptable only for crash-only services.
- Bounded resources — MemoryMax and CPUQuota on every unit; unbounded services are findings, not workloads.
- Health surface — every service exposes a health endpoint the overwatcher can probe without side effects.
- SLO arithmetic — error-budget burn is computed against declared nines; burn >70% of budget warns, breach pages.
- Least privilege — dedicated User=, no root, secrets via environment files never unit args.
Operating model
Overwatchers run read-mostly: probe health, evaluate policy, restart within budget, and escalate only on budget breach or unknown failure signatures. The checker on this page is the same rule set the runtime enforces — deterministic, explainable, no silent actions.
Cluster
Sibling properties in the MCP/A2A protocol bridge and guardrail runtimes.
MCP Context ProMCP↔A2A HubA2A→MCP CompilerOverwatcherTypedBotsStrictBotsRunWeights · ForwardRouterReverseHarnessLinkProducers