Hermes is an open-source agent runtime. It runs on your machine, holds memory across sessions, uses tools on your behalf, and can operate on a schedule without you watching.
This guide walks through what Hermes is, how it works, and how to design an agent team for a real workflow — from a single agent doing keyword research to a full team that researches, writes, and reviews content on a weekly cadence.
No coding experience required. Each chapter builds on the last.
Chat tools lose context between sessions and can't act on their own. A persistent agent runtime changes what your AI can do — and when it's overkill.
A chat app responds to prompts. An agent runtime runs tools, holds memory, and operates on a schedule. A better runtime does not fix a bad workflow.
An agent is not just a model — it's a role with identity, tools, memory, and skills. The hiring metaphor: you hire a specialist, not someone smart.
What runs underneath: Python, the ~/.hermes directory, model providers, and the tool registry. You don't need to understand most of this — but knowing it exists helps when something breaks.
The CLI is the primary interface. The gateway connects your agent to 20+ messaging platforms. Some services run continuously, others only when you ask.
Installing a runtime is different from signing up for a SaaS — your data stays on your machine. The install command, the setup wizard, and choosing between local, VPS, or Docker.
The agent loop: observe, think, act, repeat. How tool calls happen, when approvals kick in, and what this means for how you design your agents.
What makes one agent better than another: narrow roles, the instruction hierarchy, skill documents with output contracts, and feedback loops that compound over time.
Profiles for durable roles that compound over time, subagents for one-off tasks, and kanban coordination to keep agents from stepping on each other.
What agents store, when they retrieve it, and why memory is not intelligence. Auto-learning means persisting your feedback — not independent improvement.
Scheduled tasks, no-agent scripts, webhook triggers, and the boundary between what to automate and what to keep manual.
The capstone: start small with two agents, add complexity only when you see the gaps. PR-only publishing, the minimum viable setup, and the mature team design.
Security is a practice, not a feature. Secrets handling, approval workflows, provider fallback, backup strategies, and the six habits that prevent most problems.
Chapter 1 starts with the problem — why your current AI chat sessions lose context and can't carry work forward — and shows what changes when your agent runs in a persistent environment.
Book a Call