Skip to content

Verify a HermesAgent is healthy

This guide shows you the four checks that confirm an agent is fully working end-to-end, from "the CR landed in etcd" to "a real message round-trip succeeded."

Run these in order. If any check fails, the next one is meaningless until you fix it. For decision-tree-style triage, see Troubleshoot an agent that isn't Ready.

1. CR-level: status.phase is Ready

kubectl -n hermes get hermesagent my-agent

NAME       PHASE   IMAGE                                            AGE
my-agent   Ready   docker.io/nousresearch/hermes-agent:v2026.4.30   2m

If PHASE is anything other than Ready, jump to Troubleshoot an agent that isn't Ready.

Also check the conditions:

kubectl -n hermes get hermesagent my-agent -o jsonpath='{.status.conditions}' | jq

Expect PodReady=True, SecretsResolved=True, RBACSynced=True. If the dashboard sidecar is enabled, also expect GatewaysReady=True. See Lifecycle, phases, and conditions for what each condition means.

2. Pod-level: the pod is Ready and the gateway is running

kubectl -n hermes get pod -l app.kubernetes.io/instance=my-agent

NAME                            READY   STATUS    RESTARTS   AGE
hermes-my-agent-7d8c4f9b5-xqz4t 1/1     Running   0          90s

READY should be 1/1 (or 2/2 if the dashboard sidecar is enabled). RESTARTS should be 0 for a fresh deploy; non-zero is not always wrong, but worth a glance at logs to see what restarted.

Check the agent logs for the readiness signal:

kubectl -n hermes logs deployment/hermes-my-agent -c agent --tail=50 | grep 'Gateway'

Expect ✓ Gateway is running. The exec probe greps for this exact string.

3. Gateway-level: the messaging platform is connected

For each gateway entry on the CR, check the platform-side connection. The exact log line varies:

  • Discord: Logged in as <bot-name> after token validation.
  • Telegram: Bot @<handle> connected or similar.
  • WhatsApp: a QR-code prompt or the saved-session reuse line.

If the dashboard sidecar is enabled, the same information is in status.gateways[]:

kubectl -n hermes get hermesagent my-agent -o jsonpath='{.status.gateways}' | jq

[
  {
    "type": "discord",
    "ready": true,
    "state": "connected",
    "lastProbedAt": "2026-05-30T12:34:56Z"
  }
]

Without the dashboard, you only see this from the logs.

4. End-to-end: send a real message and get a reply

This is the only check that exercises the full path: Discord/Telegram/WhatsApp API → upstream gateway → LLM provider → reply.

From your messaging platform, send the agent a direct message:

  • Discord: DM the bot, or @mention it in a channel where it has View Channels + Send Messages permissions.
  • Telegram: search for the bot by handle and send /start or any message.
  • WhatsApp: message the bot's number.

A reply should come back within a few seconds. If it doesn't, check agent logs for a stack trace from the LLM call (auth, model name, rate limit, network). A common one: HTTP 400 ... but you passed anthropic/claude-opus-4.6 when using a non-Anthropic provider. See Change the LLM provider for an agent.

Optional: dashboard /api/status walk-through

If spec.dashboard.enabled: true:

kubectl -n hermes port-forward svc/hermes-my-agent-dashboard 9119:9119
curl -s http://localhost:9119/api/status | jq

{
  "gateway_running": true,
  "gateway_platforms": {
    "discord": { "state": "connected", "error_message": "" }
  }
}

gateway_running: true AND every platform state: connected means the agent is fully operational.

See also