AI Compass / Knowledge cards
Long-running jobs that wake the model only on change
wake gate
A long-running job runs on a schedule, for example every 15 minutes. On Linux, cron, the built-in scheduler, does this. If the job calls a language model every time, it costs tokens even when nothing has happened. A wake gate puts a simple check script in front. It decides by itself whether anything has changed and wakes the agent only then. AI LABS shows this with the Hermes Agent, but the pattern does not need Hermes.
What it is for
You pay for tokens only when something really happens.
Steps
- Choose a check task that repeats regularly, for example cost spikes or new reviews.
- Write a check script that reads the state and decides by itself: change yes or no. In Hermes, the script reports “no” with the last output line
{"wakeAgent": false}(according to the Hermes docs, not in the video). - On “no”, nothing happens. On “yes”, the agent receives the context and reports or acts.
- For pure monitoring, for example of certificates or payment status, use no model at all.
- Send the alert to a channel you see on the go. If needed, you bring the agent in with the existing context.
- Have actions such as replies or payments carried out only after your approval. In the video they run without approval; only posts for social networks land in a review folder first. I recommend the approval requirement.
Limits
- Credentials for email or payments do not belong in an agent’s hands without approval.
- Automatic replies to prospects are risky: wrong commitments, data protection.
- The comparison with OpenClaw in the video is a claim by the channel.
- Hermes Agent has played no role in the channels reviewed since September. The pattern works with any scheduler.
Links
- Decision models: If a fixed rule in the check script is not enough, a cheap decision model can answer the yes/no question.
- Pre-sorting your inbox: a typical long-running job that wakes only on new mail.
- Model choice: cost per task: The cheapest model use is the one that does not happen at all.
Sources
- AI LABS: 6 Insane Hermes Agent Use Cases That You Need Right Now, 10 June 2026, 2:09-4:49, 10:09. Sponsored by Brilliant (affiliate link), plus a paid community
- Hermes Agent: Cron docs, retrieved 11 October 2026
Status
Checked 11 October 2026 · Maturity new (pattern can be rebuilt with Linux’s built-in tools, not tested by me)
Want to try this with your team? That’s what the AI Lab is for. See the AI Lab