Cut Your Hermes Agent Token Bill: 10 Tips, Every Prompt
Ten ways to make Hermes Agent use fewer tokens without making it worse: see where the tokens go, start fresh chats, hand big reading to a subagent, switch off background work, prune old tool results, trim tools, skills and your SOUL, move small jobs to a cheaper model, and stop waking your agent for nothing. Every prompt from the video, in order.
16 prompts · 10 parts · copy-paste ready
// Contents
Tip 01 / 10 — See where your tokens go
Tip 02 / 10 — Start a fresh chat when the topic changes
Tip 03 / 10 — Let a helper agent do the big reading
Tip 04 / 10 — Switch off the work you never asked for
Tip 05 / 10 — Let Hermes clean up old tool results
Tip 06 / 10 — Turn off the tools you don't use
Tip 07 / 10 — Keep your always-on text short
Tip 08 / 10 — Clear out the skills you never use
Tip 09 / 10 — Give the small jobs to a cheaper model
Tip 10 / 10 — Don't wake your agent for nothing
Prerequisites
- ▸Hermes Agent installed and answering (Telegram or terminal chat)
- ▸Words in [square brackets] in the prompts are yours to fill in
See where your tokens go
See what goes out with every message
your Hermes agent
Type these straight into the chat, one at a time:
/contextSee what the whole session used
your Hermes agent
Show full prompt
/usage/context shows what goes out with every message, piece by piece. /usage shows the whole session: how many calls to the model it took, and how many tokens it read.
Start a fresh chat when the topic changes
Start a new session
your Hermes agent
When the topic changes, start a new session:
Show full prompt
/newCompress a long chat
your Hermes agent
Your agent keeps its memory and instructions. If a chat has to stay long, type this at a good stopping point instead. It costs one call to the model, so it pays off on a long chat, not a short one:
Show full prompt
/compressLet a helper agent do the big reading
Hand the research to a subagent
your Hermes agent
Give big research jobs to a subagent, so all the reading stays out of your chat:
Show full prompt
Don't do this research yourself. Hand the whole job to one subagent, and give me back just its
5 bullet points: [YOUR RESEARCH JOB]The example from the video: "find the 5 most-viewed YouTube videos about Hermes Agent from the last month, get their transcripts, and work out what their first 30 seconds have in common." The helper works in the background, so wait for its answer.
Switch off the work you never asked for
Find the background jobs
your Hermes agent
First, see what runs behind your messages:
Show full prompt
Which background jobs do you run on your own after my messages, like reviewing memory or making
chat titles? And how many model calls have they made on this machine so far?Switch the background jobs off
your Hermes agent
Then switch them off:
Show full prompt
Switch off the background review that runs after every turn, and the automatic chat titles.
Change the settings themselves, not a memory note.Your agent still remembers what you tell it to remember, and you can name a chat yourself.
Let Hermes clean up old tool results
Turn on auto-prune
your Hermes agent
Show full prompt
Turn on auto-prune for old tool results in our chats, starting at 48,000 tokens.The setting is off until you give it a number. It never touches your messages or the agent's answers, and it costs no extra call to the model.
Turn off the tools you don't use
Find the tools you don't need
your Hermes agent
Show full prompt
Which of your tools and add-ons have we never used, and which don't I need for [YOUR JOB]?
Give me a numbered list.Switch those tools off
your Hermes agent
Then tell it which ones to switch off, by their numbers:
Show full prompt
Switch off numbers [X, Y, Z].Check your other agents
each of your other agents
If you run more than one agent, check every one of them. Paste this into each other agent:
Show full prompt
You're my [ROLE] agent now. You only [WHAT IT DOES]. Which of your tools don't you need for
that job?Keep your always-on text short
Move long instructions into skills
your Hermes agent
Show full prompt
Move any long, task-specific instructions out of your SOUL into skills, and keep your SOUL short.A skill only sends its name and one line until your agent needs it.
Clear out the skills you never use
Find the skills you never use
your Hermes agent
Show full prompt
Which of your built-in skills have we never used, and which won't I need for [YOUR JOB]?
Group them.Switch those skills off
your Hermes agent
Then:
Show full prompt
Switch off those groups.Give the small jobs to a cheaper model
Put background jobs on a cheaper model
your Hermes agent
Show full prompt
Put your background jobs on a cheaper model. Keep the main chat on the model it uses now.Then type /new so the change kicks in. If a helper or a scheduled job needs real thinking, keep that one on the strong model.
Don't wake your agent for nothing
Only wake the agent when something is new
your Hermes agent
Show full prompt
Change the [JOB NAME] job so it only wakes you up when [something new has actually arrived].The agent writes a small check that runs first. If there's nothing new, the model isn't called at all, so an empty run costs nothing.
All 16 prompts are . Or skip the setup:
Membership
Build beyond this tutorial
- 3 paid tutorials of your choice today
- Every new tutorial, the day it drops
- One-line auto-installers for what you unlock
- Private members Telegram
- Priority inbox, reply within 24h
$10/month. $20 covers your first 2 months.
Become a memberGet all 16 prompts by email
No card needed.
Want them as a single Markdown file? We'll email you a 6 digit code to confirm.
// What you're building toward
An agent that spends its tokens on your work: short always-on text, only the tools it needs, background jobs on a cheaper model, and nothing running when there is nothing to do.
Stuck on a step? Drop a comment on the video, or book a session and we'll walk through it together.
Usage & License
Use the prompts and downloadable templates on this page in your own personal and commercial projects. What's not allowed: reselling, repackaging, redistributing, or republishing this material — on YouTube, Gumroad, paid courses, newsletters, or anywhere else. Send people to komputermechanic.com instead. Violations get reported and taken down.