Hermes Cron Jobs: Memory Update Guide (2026)

Wondering how to use Hermes cron jobs now that they finally have memory? The quick version is just below: as of Hermes Agent v0.21.0 (released 31 August 2026, per the official GitHub release notes), scheduled jobs retain persistent memory between runs — the release notes literally say “Your 9am briefing job now knows what it told you yesterday.” I’ve been scheduling agents in Hermes for months, and this update quietly fixes the single most annoying thing about them, so here’s the full breakdown: what changed, how the cron layer works, and how to use it without burning tokens.

Hermes cron jobs explained: what they are and what changed

Hermes Agent has had scheduling for a while — the official documentation describes it as “a built-in cron layer [that] lets you schedule any agent task and pipe the result to any messaging channel”. Morning briefings, competitor monitoring, inbox triage, recurring SEO checks: anything you’d ask the agent to do once, you can ask it to do on a schedule.

The problem, historically, was amnesia. Every scheduled run started cold. Your 9am briefing didn’t know what yesterday’s 9am briefing said, so you got repetition, re-summarising, and no sense of progress. That’s the thing v0.21.0 fixed on 31 August 2026. Straight from the GitHub release notes:

• Persistent memory between runs — “Scheduled agents now retain persistent memory between runs, enabling continuity.” The release notes put it perfectly: “Your 9am briefing job now knows what it told you yesterday.”
• Skip-when-unchanged — jobs can skip LLM calls when nothing changes. A monitor that finds nothing new just… finishes. No token spend, no noise.
• Output lands where you talk — cron output “can land in a bot’s canonical chat where it responds”. Combined with the new Bot Mode society, your scheduled job effectively becomes a named teammate that reports in and can discuss its own report.

This shipped as part of a huge release — roughly 5,800 commits and 2,475 merged pull requests since v0.20.0, per the release notes — but the cron memory change is the one that alters how you should design your automations.

How to use Hermes cron jobs properly after v0.21.0

First, update. Existing installs: hermes update. Fresh installs use the official script from the NousResearch/hermes-agent repository:

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

Then rethink your scheduled tasks around continuity, because the design rules have changed:

Before v0.21.0	After v0.21.0
Every run starts cold — no context from previous runs	Jobs retain persistent memory between runs
Briefings repeat what they already told you	“Your 9am briefing job now knows what it told you yesterday”
Every run costs an LLM call, even when nothing changed	Jobs can skip LLM calls when nothing changes
Output piped to a channel as a one-way report	Output can land in a bot’s canonical chat, where the bot responds

Practical patterns that only work now:

• Delta briefings — ask the job to report only what’s new since its last run, instead of a full daily dump. Before v0.21.0 this was impossible; now it’s the obvious default.
• Cumulative monitors — a rankings or mentions monitor that tracks how things trend across its own run history, not just a snapshot.
• Quiet-by-default watchers — pair the memory with skip-when-unchanged and you get watchers that only ever speak when something moved. Zero cost, zero noise on quiet days.

Where Hermes cron jobs fit in the bigger v0.21.0 picture

The cron upgrade isn’t standalone — it’s one leg of the release’s push towards agents that behave like a team rather than a tool. The same release notes cover Bot Mode (a society of named agents with Discord-style group chats and hermes peer messaging between profiles), live subagent steering through the delegate_task tool (list running children, correct course mid-task, stop early while keeping partial results), a unified MCP command centre with background health checks, and the agent driving the desktop’s in-app browser directly.

The combination matters for anyone doing SEO or content work with agents. A scheduled job with memory feeding into a bot’s chat, where you can steer follow-up work live, is a genuinely different workflow from “script fires, dumps text, forgets everything”. I’ve written before about Hermes Bot Mode and subagent steering — cron memory is the third piece that makes both of those click. And if you’re tuning how much the agent remembers generally, my guide to Hermes memory setup covers the wider memory system these scheduled jobs now plug into.

One honest caveat: the release notes describe what the features do, but your token costs and results depend entirely on how you design the jobs. A sloppy prompt on a schedule is just sloppy output on a schedule. This is exactly the kind of thing we workshop on the live calls in the Boardroom — bring your job prompt, leave with a better one.

The bottom line on Hermes cron jobs

Hermes cron jobs went from “cron that runs an AI” to “a scheduled teammate with a memory” in one release. Update to v0.21.0, redesign your recurring jobs as delta reports rather than full dumps, switch monitors to skip-when-unchanged so quiet days cost nothing, and pipe output into a bot chat so you can actually talk back to your automation. That’s the play. If you want the exact recurring-job setups we use for SEO — content refreshes, rank monitoring, briefing agents — they’re inside the AI Profit Boardroom, or book a free SEO strategy session and I’ll help you plan yours.

FAQ: Hermes cron jobs

What are Hermes cron jobs?

They’re Hermes Agent’s built-in scheduler. Per the official docs, “a built-in cron layer lets you schedule any agent task and pipe the result to any messaging channel” — daily briefings, monitors, recurring content tasks and so on.

Do Hermes cron jobs remember previous runs?

Yes, as of v0.21.0 (released 31 August 2026). Scheduled agents retain persistent memory between runs, so a job knows what it already told you and can build on it instead of repeating itself.

Do Hermes cron jobs cost tokens on every run?

Not necessarily. The v0.21.0 release notes state that jobs can skip LLM calls when nothing changes, so a monitoring job that finds nothing new can complete without spending anything.

Where does Hermes cron job output go?

You can pipe results to any messaging channel, and since v0.21.0 output can land in a bot’s canonical chat in Bot Mode, where the bot responds in context like a teammate.

How do I get the new cron memory features?

Update to v0.21.0 or later. Existing installs run hermes update; fresh installs use the official install script from the NousResearch/hermes-agent GitHub repository.

How are cron jobs different from Hermes Bot Mode?

Bot Mode is the multi-agent society — named bots, group chats and peer-to-peer messages. Cron jobs are the scheduler. They combine: a scheduled job can deliver into a bot’s chat, giving you a bot that proactively reports on a schedule.

Related reading

• Hermes Bot Mode: the multi-agent society explained
• Hermes subagent steering
• Hermes memory setup

Last updated September 2026. This is the living guide to Hermes cron jobs — it gets updated as the tools change.