Read a Quick Bite when you want a short explanation of the tools we discuss.
Digital Garden / Quick Bites
Why I push Hermes?
I push Hermes because I want more say in how my tools work. I like being able to choose among supported AI providers, customize the setup, and keep useful instructions as reusable skills instead of explaining the same task from scratch. I also want people to understand enough to make changes themselves. Start with one useful job, learn how the pieces fit, and build from there. That is the kind of control I care about: knowing what you can change, what you still depend on, and when you need help.
Hermes documents skills, memory and provider choices. Those capabilities appeal to this approach, but more control also means more settings, permissions and upkeep to understand. We start with what you already use and decide whether a change would help.
Hermes Agent is open-source agent software from Nous Research. It connects a selected AI model to tools, instructions and work context. It is not itself a replacement name for the model.
A skill is a reusable procedure with instructions and, when needed, supporting files. Memory and saved skills can help carry context forward, but they do not remove the need to check results or approve access.
You can run an agent in different environments. What it can read, change or send depends on the configured tools, permissions and provider. An agent’s suggestion is not your approval to change or publish anything.
Tailscale creates a private network between enrolled devices, called a tailnet, using encrypted connections. In a learning example, your phone can reach a small app running on your laptop without treating that app as a public website.
Access rules, device security and account protection still matter. An optional exit node routes internet traffic through a chosen device; it is a different job from simply connecting two devices.
The official pricing page offers a free Personal plan for non-commercial use. Business use needs an eligible commercial plan. Check current terms, limits and features before choosing the setup.
Cloudflare: a front door, not a substitute for permissions
Cloudflare provides services between visitors and an application. Cloudflare Tunnel uses an outbound connection from a server to Cloudflare rather than requiring a publicly routable address on that server. A tunnel alone is not a sign-in policy; authentication and access restrictions need separate configuration.
A private tailnet and an authenticated web address solve different access needs. You might use the former for your devices and the latter for invited users. Neither arrangement should be described as automatically safe for every kind of data.
visitor → access checks → web app
↓
private database
staging → review → production
worker → approved API → model
This is a generic illustration, not a client deployment. A website on a VPS, a database, a voice agent and an automation each need their own access, hosting and failure-handling decisions. A staging copy is checked before replacing the public service.
Use this starter to check what your AI tool can access, choose one safe task and decide how you will verify the result. Download it and adapt it to your work.
# ETC DIY starter
status: draft-template
scope: low-risk practice only
## TITLE
[Name one small task]
## CHECK
Ask me: What result do I want? What do I use today?
Can this interface read local files or only uploads?
Can it run tools, or only explain steps?
What accounts, connectors and paid services are needed?
What information must stay out? What requires approval?
Ask me to explain the next step back in my own words.
## BODY
Goal: [one observable result]
Inputs: [non-sensitive examples]
Steps: [review before running]
Verify: [how I check the result]
Stop: [when to ask for help]
Recover: [how to undo or restore]
## LIMITS
Do not activate spending, publish, send messages or
change permissions without my explicit approval.
Never treat a confident answer as verification.
A readable file can help a person or an agent follow instructions. An llms.txt file does not guarantee indexing, use or recommendations.