4 local Hermes agent workflows I still use after a month

Four workflows I keep running on a local Hermes agent — the ones that survived a month of daily use instead of being set up once and left to gather dust. They run on a Mac mini with a local model, so there’s nothing to pay per run.

The four workflows

1. Morning brief. Every morning at 6:45, my to-dos, reminders (mine and my kids’), and the day’s key trends land on my phone through Telegram — without opening an app or writing a prompt. Keep the brief short; a wall of text is the fastest way to stop reading it.
2. Inbox to to-do. Instead of a plain summary, the agent turns my email into what actually needs doing — a short list of the most urgent plus the most important — pushed to my phone before I’m even up.
3. Keep watch. Hand the agent a URL and it tells you only what changed — a competitor’s pricing page, a supplier’s stock, your own product page — rather than dumping the whole page back at you.
4. Just talk. Capture things by voice, even replying from an Apple Watch, so jotting down a thought feels like handing it to an assistant standing next to you.

Setup gotchas that trip people up

• Pick Telegram as the messaging channel. I tried WhatsApp, LINE, and WeChat; Telegram was the simplest and smoothest to set up.
• Set the machine’s sleep to never. Otherwise the scheduled job won’t run when the time comes and the computer is asleep.
• Get your timezone right, or the brief arrives at the wrong hour.

When local is the right call

Under the hood the four are really two patterns: a job that repeats on a schedule, and “fetch, then summarize.” Both are a great fit for a slow-but-free local model — it can grind away while you sleep, so speed barely matters. If you’re already on a paid monthly plan, running these through the cloud would burn tokens; Claude’s Cowork can draw on your subscription instead.

New to Hermes? Start with the full local setup: Build a local AI agent with Ollama + Hermes.