Skip to content
vavio

My Hermes job ran, so why did no result arrive?

A Hermes scheduled job that ran but sent you nothing usually did one of four things: it failed to deliver (last_status: delivery_failed), it was told to stay quiet ([SILENT], an empty script-only result or a pre-check that didn’t wake the agent), it produced a blank reply, or its failure notice was held back because you were already alerted. The job’s status, its three error fields and its saved output tell these apart. Start with the job record, not the chat you expected the message in.

This follows Hermes Agent v0.23.1, commit 3b42157, checked on 16 November 2026. We reviewed official documentation and source; we have not run a live delivery trial on a Vavio computer. The commands below are upstream Hermes commands, not Vavio CLI commands.

Which case am I in?

Ask Hermes in chat to list your scheduled jobs. The job list includes each job’s delivery target, last_status, last_error, last_delivery_error, last_fire_error and last_delivery_unverified. If you have the Hermes command line, hermes cron list shows the same facts in words.

What the job record shows What happened Where to look next
last_status: delivery_failed with last_delivery_error The run succeeded; sending its result failed. The delivery target and the error reason.
last_status: ok, no errors, nothing arrived The reply was silent, the target is local, or the target is origin with nowhere to send. The job’s delivery target and its saved output.
last_status: error with “Agent completed but produced empty response” The agent finished with a blank reply. The model, its limits and the run’s saved output.
last_status: error with another last_error, no notice in chat The run failed and its notice was held back or already sent earlier. hermes cron incidents and the run history.
last_fire_error set, or the next run shown as Overdue The scheduled run never started. The gateway, not the job.
⚠ Delivery UNVERIFIED The platform accepted the send without proof it landed. The destination chat itself.

hermes cron doctor runs these checks for every active job at once and changes nothing.

The result was sent but never arrived: delivery_failed

Hermes records execution and delivery separately. When the agent succeeds but the message doesn’t reach the target, the job’s last status is delivery_failed, never ok, and the reason is in last_delivery_error. Upstream lists platform server errors, rate limits, stale sessions and an adapter that doesn’t report a successful send as causes. A send through a platform’s fallback sender that takes longer than its timeout (60 seconds by default) is also recorded there, and the message may still land later.

The job did its work: a delivery failure doesn’t count toward its failure streak, and the output is still saved. Fix the target, then check the next run. hermes cron edit <job_id> --deliver <target> changes where results go.

Delivering to all sends to every connected home channel. If no home channel is configured when the job fires, it has nowhere to send, and upstream records that as a delivery failure.

The run succeeded, but Hermes stayed quiet on purpose

Several settings suppress delivery for a successful run while still saving its output in the Hermes home’s cron/output/<job_id>/ folder:

  • The reply counted as silent. A reply is silent when the whole reply, its first line or its last line is [SILENT], SILENT, NO_REPLY or NO REPLY (ignoring case), or when it starts with [SILENT]. A report that mentions [SILENT] in the middle of a sentence is still delivered. If your prompt asks for [SILENT] when nothing changed, a model that appends “SILENT” as its own last line to a real report will suppress that report. Open the saved output to see what it actually said.
  • A script-only job printed nothing. For a no_agent job, empty output is a deliberate quiet tick. A non-zero exit or a timeout still sends an error alert.
  • A pre-check decided not to wake the agent. A script that ends with {"wakeAgent": false} skips the agent for that run, so there is nothing to deliver. A monitor job whose watched output hasn’t changed is quiet in the same way.

Check the delivery target too. A job created from the standalone CLI defaults to local: its results go only to the saved output folder. Upstream documents origin, meaning back to the chat where you created the job, as the default for jobs created on a messaging platform. If origin has no chat to return to and no home channel is configured, Hermes treats it as local rather than as an error. hermes cron list prints the target as Deliver.

Hermes finished with a blank reply

In the pinned source, a run whose final reply is empty or whitespace is recorded as a failure, with last_error set to “Agent completed but produced empty response (model error, timeout, or misconfiguration)”. Nothing is sent for that run: Hermes decides what to send from the reply before it records the blank reply as a failure, so there is no message and no failure notice.

The job still looks broken in the job list and its failure streak goes up, so the record tells you even when your chat doesn’t. Look at the model the job uses, its rate or usage limits and the run’s saved output. Why Hermes rejects a model or ignores its tools covers model-side causes.

The run failed, but the notice was held back

Failed runs normally send a failure notice even if the prompt asked for silence. These exceptions explain a failure you never heard about:

  • Same error, already alerted. Hermes records repeated failures with the same error as one incident. The first failure alerts you. Repeats stay quiet, with one reminder each time cron.failure_repeat_alert_hours passes (6 hours by default). A different error alerts immediately. Run hermes cron incidents to see them.
  • You acknowledged it. hermes cron incidents ack <id> silences that exact error for that job for good, even after later successful runs. A different error still alerts. The runs and the failure streak are still recorded.
  • Failure notices go somewhere else. A job can have a separate failure target; one set to local sends no failure notices.
  • An automatic re-run is pending. When a recurring job fails on a network or DNS error before reaching the model, Hermes re-runs it after 5, 15 and 30 minutes and holds the interim notice. This is on by default (cron.retry_unreachable), and a re-run never fires later than the job’s own next scheduled run, so a job that runs every few minutes alerts normally. You get the real result when a re-run succeeds, or one failure alert once the re-runs are used up.

A job that finishes its turn but reports a failure in prose stays successful. If the agent puts [CRON_FAILURE] alone on the first line of its reply, the run is recorded as failed and its explanation is delivered as a failure notice.

The job never actually ran

If hermes cron runs <job_id> shows no attempt at the expected time, the problem is upstream of delivery. A next run marked Overdue usually means the scheduler stopped ticking or the gateway is down; upstream’s fix is restarting the gateway. On hosted deployments where an outside scheduler triggers the gateway, a failed hand-off is stamped as last_fire_error. The usual sign is a job that works when you run it manually but never fires on its own. Why a Hermes scheduled job runs at the wrong time covers clock and catch-up rules.

The platform says it sent, but you can’t find it

Hermes logs a delivery as delivered only when the platform returns evidence such as a message ID. Some adapters (upstream names Slack, Matrix and Mattermost) acknowledge without that evidence; the job is then marked Delivery UNVERIFIED rather than failed. Look in the destination itself.

On Telegram, check notification settings before assuming the message is missing. Cron results are sent with a push notification by default. If cron.delivery.notify is set to false, they arrive silently under Telegram’s default notification mode, and upstream notes that people report silent briefs as never delivered.

What should I ask Hermes?

List my scheduled jobs. For “Morning brief”, show its delivery target, last run time, last_status, last_error, last_delivery_error and last_fire_error. Then open its most recent saved output and tell me whether the reply counted as silent.

Re-check after the next scheduled run rather than after a manual one. A manual run tells you the job works; only the next scheduled run shows the schedule and delivery path together.

Vavio supplies a managed remote Hermes computer for $14.99 per month, with model usage billed separately. A hosted gateway keeps scheduled jobs running while your laptop is off; a job’s record still tells you whether its result reached you. See the current hosting offer, or use these upstream checks on your own server if you prefer to maintain it yourself.

Sources

Written by Vavio with AI assistance and checked against pinned official Hermes v0.23.1 documentation and source. This is a source review, not a live delivery trial.

More from the Vavio blog: Hermes Agent guides