🌱 Welcome to Factnetize — honest, well-researched articles on tech, health and AI. Get the weekly →
Agentic AI

Hermes cron not working: every message and its fix

A Hermes cron job that never fires or never arrives is rarely broken itself. Look up the exact message from hermes cron status, see what it means, and apply the fix.

A white gateway device with a green status LED on a dark desk beside a spiral calendar, standing in for a Hermes cron job that did not run
✓ Verified against Hermes Agent v0.23.2. Every command, setting and docs link in this article still exists in Hermes.

⚡ Key takeaways

  • Hermes cron jobs are fired by the gateway, which ticks the scheduler every 60 seconds.
  • hermes cron status is the first check: it says in one line whether your jobs will fire.
  • A job never fires, is blocked before the model, or its result never arrives. Most fixes are one command: hermes gateway restart.

Hermes cron not working? A job that should have run this morning and did not is rarely broken itself. Usually the scheduler stopped ticking, a check blocked the run, or the result went nowhere. This page lists the messages Hermes shows for each case, word for word as they stand in the Hermes source code on November 21, 2026, with the fix for each. If your job does run but fails because of its model or prompt, read Why your Hermes cron jobs fail instead.

First check: is the scheduler ticking?

The Hermes cron docs are clear: cron execution is handled by the gateway, which ticks every 60 seconds. In Hermes Desktop, the cron panel marks a job that should have run with Overdue since. For the cause, open a terminal:

  1. Run hermes cron status.
  2. Look for your exact message below.
  3. Apply the fix, then run hermes cron status again.

The good answer is ✓ Gateway is running — cron jobs will fire automatically. Is that what you see, and the job still did not do its work? Then skip to blocked before the model or the job ran, but nothing arrived: those messages come from the alert and from hermes cron list.

Hermes cron not working: the job never fires

“✗ No scheduler is serving profile ‘default’ — its cron jobs will NOT fire”

No gateway is ticking this profile. Start it with hermes gateway start. Run hermes gateway install once so it comes back at login; on Windows that creates a Scheduled Task. Our Hermes Telegram not working guide covers the gateway itself.

“⚠ Next run <time> is OVERDUE — passed <N>h ago but the job has not fired (is the scheduler running?)”

A next run stuck in the past is the signature of a scheduler that stopped ticking. Hermes only calls it OVERDUE after 15 minutes, so this is a real gap. Run hermes gateway restart, or run the job now with hermes cron run <job_id>.

“⚠ Gateway is running but the cron ticker has not reported a heartbeat.”

The gateway is up, but its ticker has not reported in. If the gateway just started, wait about 60 seconds and check again. If the heartbeat never appears, run hermes gateway restart.

“⚠ Gateway is running but the cron ticker looks STALLED — no heartbeat for Ns (expected every ~60s).”

The gateway process is alive, but its ticker has gone quiet. Jobs may not be firing. Run hermes gateway restart.

“⚠ Gateway is running STALE code — its cron ticker yields every tick and fires NOTHING.”

This follows a hermes update: the code changed under a running gateway, so the ticker gives up on every tick while it still looks healthy. Run hermes gateway restart to load the new code.

“⚠ Gateway and cron ticker are running, but no tick has succeeded”

The loop runs, and every tick fails. The next line shows the last tick error. With Permission denied, the file jobs.json may be owned by another user, for example after a root docker exec; give it back to the gateway user. With a file-descriptor error, restart the gateway.

“⚠ Cron store is NOT writable — scheduled jobs are being skipped”

Hermes cannot write to its cron store in ~/.hermes/cron/, so it skips jobs rather than risk running them twice. The lines below name the path and the error. Free disk space, remount the drive read-write, or fix the folder permissions. No restart is needed: on the next tick, each job that stayed due fires once.

Blocked before the model: “blocked_config”

Before a run starts, Hermes checks that the job can work: the provider key, the attached skills, the delivery target and the MCP servers it names. If one fails, the job’s status becomes blocked_config, one alert goes out, and no model is called, so no tokens are spent. You see it in that alert and in hermes cron list. A missing key also names the profile and the folder the scheduler read, which explains a job that fails while the same key works in your chat.

  1. Read the alert for the missing piece.
  2. Add the key, skill, delivery target or MCP server it names.
  3. Run hermes cron run <job_id> to test it.

The job ran, but nothing arrived

“⚠ The result was not delivered: <reason>”

hermes cron list shows this line under the job, and its status is delivery_failed. The agent did its work, so it does not count toward the failure streak. Check the reason, then the target: hermes cron edit <job_id> --deliver <target> changes it.

“⚠ missed scheduled fire at <time>: … The messaging gateway was unreachable.”

The scheduled run could not be handed over. The message itself gives the fix:

  1. Run hermes gateway restart.
  2. Run hermes cron run <job_id> to run the job now.

“⚠ Last failure at <time>: <reason> — recovered since”

Nothing to fix. The stamp stays after a later success, so you can see that the job failed once. “Recovered since” means the last run was fine.

Check that it works

  1. Run hermes cron status and look for ✓ Gateway is running — cron jobs will fire automatically.
  2. Run hermes cron doctor. A healthy setup prints ✓ Cron doctor found no issues.
  3. Run hermes cron list and read the lines under your job. For a job that keeps failing, hermes cron incidents lists the errors.

What’s next

If the scheduler ticks and the job still fails, the cause is in the job: its model, its prompt or a missing fallback. Why your Hermes cron jobs fail covers that side.

This article is part of the Agentic AI series.

John Lock
Written by

John Lock

The Tested Weekly

One email a week with the guides we tested, sources linked. No spam.

Leave a comment

Your email address will not be published. Required fields are marked *

Weekly · No spam

The Tested Weekly

One email a week with the guides we tested, sources linked.