← All courses

Hermes Agent: Build, Automate, and Operate AI Agents

A practical, security-first course on configuring Hermes Agent and using it for automation, development, research, and multimodal work.

Log in to start →

Course preview

What you will learn

Explore the full curriculum before you enrol. Lessons unlock after purchase or free enrolment.

01

Getting Started with Hermes Agent

Install Hermes on supported platforms, choose a model provider, and make your first chat. Compare the CLI, TUI, desktop application, and Android/Termux paths before adopting an interface for daily work.

Topics covered

  • Install the supported Hermes distribution and verify its dependencies.
  • Choose a provider that matches cost, privacy, and capability needs.
  • Start, update, and safely remove a Hermes installation.
1 lesson5 questions~25 min
02

Interfaces, Sessions, and Configuration

Manage durable sessions, configuration precedence, model settings, context files, checkpoints, managed scope, and profiles. The module turns an ad-hoc local agent into a reproducible working environment.

Topics covered

  • Resume and search sessions instead of losing working context.
  • Use configuration precedence deliberately.
  • Configure models independently from interface choices.
1 lesson5 questions~25 min
03

Models, Providers, and Inference

Design resilient inference with cross-provider fallbacks, credential pools, subscription proxying, Nous Tool Gateway, mixture-of-agents patterns, and local serving through llama.cpp or vLLM.

Topics covered

  • Separate cross-provider fallback from same-provider credential rotation.
  • Choose recovery behavior based on the error class.
  • Use a subscription proxy without leaking credentials.
1 lesson5 questions~25 min
04

Memory and the Skills System

Build persistent agent behaviour with MEMORY.md, USER.md, memory providers, Honcho, progressive disclosure skills, curation, and SOUL.md. Focus on keeping long-lived instructions useful and safe.

Topics covered

  • Distinguish user memory from project memory.
  • Select a memory provider and retrieve prior work safely.
  • Use progressive disclosure to keep tool instructions compact.
1 lesson5 questions~25 min
05

Tools, Code Execution, and Browser Automation

Use toolsets, tool search, code execution, browser automation, web search/extraction, event hooks, and vision while applying the appropriate execution limits and security boundaries.

Topics covered

  • Enable the smallest useful toolset for a task.
  • Use tool search instead of loading every MCP tool.
  • Run generated code within explicit limits.
1 lesson5 questions~25 min
06

Messaging and Platform Integrations

Connect Hermes to Telegram, Discord, Teams, email, and webhooks through a secure gateway. Learn allowlists, delivery behaviour, persistent sessions, privacy mode, and the differences between bot and webhook integrations.

Topics covered

  • Apply deny-by-default access to messaging gateways.
  • Configure Telegram and understand its group privacy mode.
  • Choose Discord session behaviour deliberately.
1 lesson5 questions~25 min
07

Scheduling, Cron, and Batch Automation

Turn repeatable work into cron jobs, watchers, batch pipelines, and daily briefings. Fresh scheduled sessions need self-contained prompts, observable delivery targets, resumability, and a troubleshooting plan.

Topics covered

  • Write self-contained prompts for fresh cron sessions.
  • Read and validate cron schedule expressions.
  • Use the SILENT pattern and explicit delivery targets.
1 lesson5 questions~25 min
08

Multi-Agent Orchestration and Delegation

Delegate isolated work, run bounded parallel batches, use durable Kanban tasks and Git worktrees, and coordinate external coding agents. The lesson emphasises review stages and the limits of subagent durability.

Topics covered

  • Choose delegation only for bounded, independently verifiable work.
  • Use parallel batches without shared-state conflicts.
  • Apply a two-stage review to subagent output.
1 lesson5 questions~25 min
09

Security, Secrets, and Deployment

Apply Hermes’ layered security model: command approvals, hard blocks, file-write safety, gateway authorization, container isolation, SSRF controls, Docker operations, egress credential proxying, and external secret stores.

Topics covered

  • Use approval and blocklist controls for dangerous commands.
  • Treat file writes and gateway access as separate security boundaries.
  • Deploy Hermes with container isolation and mandatory authentication.
1 lesson5 questions~25 min
10

Plugins, MCP, and Custom Tools

Extend Hermes through opt-in plugins, MCP servers, OAuth, mcporter, FastMCP, dashboard extensions, skins, and custom tools. Learn discovery order and design integrations as explicit, reviewable contracts.

Topics covered

  • Create an opt-in plugin with a clear manifest and lifecycle.
  • Understand plugin discovery order before debugging loading.
  • Connect to MCP with appropriate OAuth boundaries.
1 lesson5 questions~25 min
11

Voice, Vision, and Multimodal Features

Configure the complete voice loop—speech-to-text, text-to-speech, wake word, streaming, and barge-in—then add vision, image input, and generation while choosing providers and fallbacks deliberately.

Topics covered

  • Set up a low-cost local Whisper plus Edge TTS stack.
  • Compare STT/TTS providers by quality, cost, and latency.
  • Use wake-word, streaming, and barge-in behaviours correctly.
1 lesson5 questions~25 min
12

Productivity, Research, and Domain Skills

Use bundled skills for Google Workspace, Airtable, XLSX, DOCX, PDF, maps, Teams meetings, YouTube, Spotify, Polymarket, humanisation, and diagrams. The central principle is verify outputs and confirm irreversible actions.

Topics covered

  • Confirm before sending mail, changing calendars, sharing, or deleting files.
  • Use PATCH rather than destructive replacement for Airtable updates.
  • Verify XLSX formulas and DOCX/PDF rendering rather than trusting exit codes.
1 lesson5 questions~25 min
13

Software Engineering Workflows

Use systematic debugging, RED-GREEN-REFACTOR, API and language debuggers, LSP diagnostics, independent review, GitHub workflows, planning, spikes, and codebase inspection to make agent-assisted development reliable.

Topics covered

  • Find root cause before proposing a fix.
  • Run TDD as vertical RED-GREEN-REFACTOR slices.
  • Debug APIs by isolating the failing layer.
1 lesson5 questions~25 min
14

Desktop Automation, QA, and Developer Utilities

Drive native apps in the background with capture-click-verify, test web products through a five-phase QA process, create code wikis, deliver files through chat, and selectively enable bundled plugins.

Topics covered

  • Prefer background computer control to stealing a user’s cursor.
  • Capture before input and verify every state-changing action.
  • Escalate from element targeting only when evidence requires it.
1 lesson5 questions~25 min
15

MLOps, Evaluation, and Local Model Tooling

Track experiments in Weights & Biases, benchmark models with lm-eval-harness, manage Hugging Face assets, invoke hosted models via infsh, operate ComfyUI, and maintain a machine-readable DESIGN.md.

Topics covered

  • Log experiments, artifacts, and lineage in Weights & Biases.
  • Use standard benchmarks and interpret them cautiously.
  • Treat unsafe code evaluation as a deliberate security gate.
1 lesson5 questions~25 min
16

Social Platforms, Knowledge Tools, and Code Review

Use specialised skills for iMessage, a dedicated agent email identity, X content, Obsidian knowledge, academic writing, and GitHub pull-request review. Each workflow requires clear scope, consent, and a human checkpoint for external effects.

Topics covered

  • Confirm before reading or sending personal communications.
  • Keep a dedicated agent identity separate from a personal inbox.
  • Treat social content as untrusted input before acting on it.
1 lesson5 questions~25 min
17

Final Exam

Final exam
35 questions~60 min