Route each job to a cheaper model

Use case

One premium model on every Hermes turn gets expensive fast. This workflow checks that cheap jobs stay on cheaper models and that only the hard jobs use a paid one, with a hard monthly cap so a loop cannot run all night.

Edit before you send: The config path, the monthly cap, and which roles may use a paid model.

Prompt to paste into Hermes Agent

Read my Hermes model config at [CONFIG PATH, e.g. /opt/data/config.yaml]. Do not change it until I say yes. I want this routing: - Sorting, extracting, and short summaries: the cheapest model that still follows instructions. - Research notes and first drafts: a mid-tier model. - Final review, coding, and anything I marked as customer-facing: [PAID MODEL] only. - If a cheap model errors, fall back one step up, not straight to the most expensive model. - Monthly spend cap: [20] in the provider account. If you cannot see the cap, tell me to set it before we continue. Report: 1. Each role and the model it uses now. 2. Any role on a paid model that should not be. 3. A rough monthly cost if I keep running about [90] sessions. 4. The exact config lines you would change. Wait for my yes before writing them. Save a skill named cost-check that repeats this report. I will run it weekly. If session count stays flat and spend jumps, say which fallback fired.

More Hermes Agent prompts

Chief of Staff that routes the team
You have several Hermes roles and you are still deciding who should do what. This is the standing entry point: one message can mix research, a draft, an outbound note, and a scheduling task, and Hermes splits them before anyone acts.
Inbox sorted before you wake up
The first half hour of the morning disappears into overnight email. Hermes reads what arrived, separates what needs you from what can wait, and leaves reply drafts in Gmail. You open Telegram, work the urgent items, and send only the drafts you still agree with.
Scored leads with a real first line
A blast of the same cold note does not get replies. You give Hermes a tight market, and it returns a sheet of prospects with a score and an opening line taken from something it actually found about that company.
Overnight demo page for each new lead
Building a custom page for every prospect by hand does not scale. Overnight, Hermes reads the lead sheet, looks at each company site, writes a one-page demo aimed at their problem, and puts a public URL back on the row. In the morning the first email can point at something already live.