Context
The agent knows who it is, who its owner is, communication style, and which workflows to prioritize. Without context, the agent is just a generic assistant with no personality.
not an instant setup
Building an agent like Hermes takes time.
This guide explains how to build an agent that is smart, autonomous, independent, has its own accounts, its own access, and still has boundaries. This is not a short template — it is a step-by-step documentation.
What makes an agent feel independent?
Access
The agent has accounts, wallets, email, GitHub, Discord, browser, server, or tools that are set up specifically for it — not borrowed from the user. This access is what makes the agent truly able to work.
Boundaries
The agent knows when to act independently, when to just log, and when to ask for explicit permission. These boundaries are not about limiting creativity — they are about preventing risky mistakes.
SOUL.md is the center of permanent rules. Memory and corrections make the agent more stable over time. The longer the agent runs, the better it understands your preferences and workflows.
Why do you need this guide?
Many people try to make an "autonomous" agent with one long prompt, then get disappointed. A truly independent agent is not born from one prompt — it is shaped by real access, permanent rules, repeated testing, and corrections saved as memory.
This guide will take you from zero to having an agent that is truly reliable: has its own accounts, knows its limits, and gets smarter over time.
What is Waguri?
Waguri is a real example of an agent built with this approach. It is a Familiar — not a generic chatbot, not a regular assistant. It has its own accounts (X/Twitter, GitHub, Discord, email), its own wallet with full control, and can execute complex tasks autonomously.
You can see how Waguri's SOUL.md is written on the About Waguri. That page is not a template to copy — it is a real reference for how the concepts in this guide are applied.
Hermes SOUL Guide — building a smart agent is a process, not an instant prompt.