Health

The public JARVIS health endpoint — service availability, liveness, and readiness, without exposing diagnostics.

JARVIS exposes one public endpoint for service status. It requires no credential.

Endpoint

Method	Path	Auth	Purpose
`GET`	`/api/health`	None	Current service status

curl -s https://jarvis.spacekeep.dev/api/health \
  -H "Accept: application/json"

Response

{
  "status": "operational",
  "checks": {
    "core": { "ok": true, "detail": "Service responding" },
    "memory": { "ok": true, "detail": "Data store reachable" },
    "auth": { "ok": true, "detail": "Session service ready" }
  },
  "checkedAt": "2026-01-01T00:00:00.000Z"
}

Status values

`status`	Meaning	What to do
`operational`	Every check passed	Nothing
`degraded`	At least one check failed	Retry shortly; see the failing check

The HTTP status is 200 in both cases — read the status field rather than relying on the response code alone.

Checks

Check	Question it answers
`core`	Is the service responding to requests?
`memory`	Can the service reach the data it needs to answer and to recall memories?
`auth`	Is the session service ready to authenticate you?

Each check is a boolean with a short human-readable detail. The endpoint reports availability only — no metrics, no configuration, no secrets.

What it does not tell you

Which specific component behind a check is failing.
Capacity, latency, or throughput.
Anything about your account.

If you need detail, that is deliberate — this endpoint is public and unversioned. Raise an issue with SpaceKeep Labs support instead of probing for diagnostics.

Using it

As a liveness probe

const res = await fetch("https://jarvis.spacekeep.dev/api/health", {
  headers: { Accept: "application/json" },
  signal: AbortSignal.timeout(5000),
});
const { status } = await res.json();
if (status !== "operational") process.exitCode = 1;

As a readiness probe

curl -sf https://jarvis.spacekeep.dev/api/health \
  | grep -q '"status":"operational"' \
  && echo ready \
  || echo not-ready

Before you depend on it

Give your probe a timeout. Individual checks are bounded, but a slow network should not turn a health check into a hang.

Responses are sent with Cache-Control: no-store, so a probe never reads a cached answer.

In the app

The same status is shown in the interface — on the landing page, on the Developer page, and in the dashboard. You do not need to call the endpoint to check.

API Overview — the rest of the API
Global Enforcement — the difference between degraded and restricted
Troubleshooting — what to do when things are not working