The cheapest way to know whether jaque is worth your time is to run it against a machine you own and watch it decide, on its own, whether anything is wrong. This page gets you from the binary to that first decision: one host in CUE, one run, and the three HTTP surfaces every engine serves without being asked. It assumes nothing beyond the binary and a network address to ping.

Status: there is no public release and no public repository during the pre-alpha; the binary is handed directly to people granted access. Everything below is exercised by the project's own tests and dogfooding. See Status.

1. Get jaque

You start with the binary: one static executable and nothing else to install. There is no build step and no runtime dependency to satisfy first. The event log, the state machine, the scheduler and the dashboard are compiled into it.

2. Write a config

A Nagios host, once the templates are stripped away, is an address and a check. jaque's CUE config is that shape with nothing added. Save this as jaque.cue:

hosts: gw: {
	address: "192.168.1.1"
	check: {type: "icmp", host: address}
}

This is a complete, loadable configuration: one host named gw, checked by ICMP every 60 seconds, three consecutive failures before the problem goes hard. Both numbers are schema defaults, filled in by CUE, not by code that guesses. Point address at something on your own network, or at 127.0.0.1 to check the machine you are sitting on.

The reason a config this short loads is that the schema is closed and complete: every field you did not write has a default, and every field you misspell is an error naming the field. Why CUE is the argument; CUE in practice is the walkthrough.

3. ICMP needs one sysctl

The icmp check opens an unprivileged datagram socket rather than a raw one, so jaque never needs root. Most Linux distributions ship with that socket type disabled for every group, and the symptom is not an error at startup but an UNKNOWN result with "permission denied" on the first check. Enable it once:

sudo sysctl -w net.ipv4.ping_group_range="0 2147483647"

A narrower group range works as well if you prefer it.

4. Run it

jaque -config jaque.cue

The engine writes one structured log line per state transition. Nothing is notified: a fresh config wires no contacts, and jaque never invents a recipient. What you see is the engine deciding OK, WARNING, CRITICAL or UNKNOWN for each object as results arrive, and walking the soft attempts before it commits to a hard state. State model is the rulebook for those lines.

5. What you get for free

With the default -listen 127.0.0.1:8080, the same process serves three things on one port:

The dashboard at http://127.0.0.1:8080/: every host and service, its current state, and the actions an operator reaches for first (acknowledge, downtime). See Dashboard.

/metrics: Prometheus exposition format, about the engine itself rather than your hosts -- event log lag, scheduler drift, check durations. Scrape it with anything that speaks the format. See Metrics.

/status.json: a machine-readable snapshot of every object, the same shape -snapshot-out writes at shutdown. Its first use is a liveness probe; see Status JSON.

Set -listen "" to turn all three off, which is what a process running as a pure engine or worker target does anyway.

6. Where to go next

CUE in practice adds services, shares a template across hosts, and wires contacts so a hard state reaches someone. Concepts explains the state machine that produced the log lines you watched. If the config you care about already exists as a Nagios tree, From Nagios compiles it instead of asking you to retype it.

7. Security considerations

The default -listen address is loopback. Binding it anywhere else exposes the dashboard, /metrics and /status.json to that network with no authentication of their own; put a reverse proxy or a network policy in front first, as Security describes. The ICMP sysctl grants every process in the range the ability to send echo requests, which is the least privilege that makes the check work.