Skip to Content
View all events

Build Your First Local Agent with Hermes

Published by O'Reilly Media, Inc.

Intermediate content levelIntermediate

A hands-on introduction in two hours

What you’ll learn and how you can apply it

  • Understand the local AI agent landscape
  • Pick the right local model for your hardware
  • Install and run Hermes Agent end-to-end
  • Use Hermes’s persistent memory and auto-generated skills
  • Apply a decision framework for when to use Hermes Claude Code, or OpenClaw based on your workflow

Course description

AI agents are moving from the cloud to your own laptop. Local models like Qwen 3.6, Gemma 4, and Llama 4 are now capable of powering real coding and automation agents privately, cheaply, and without sending your data anywhere.

In this hands-on 2-hour course, you’ll get an introduction to the local AI agent landscape and build your first working local agent using Hermes Agent, an open source framework from Nous Research that has grown to 140,000+ GitHub stars in three months. Lucas Soares starts you off with the local models that actually work well, which one best fits your hardware, and how Hermes compares to other local coding agents like Claude Code (against Ollama) and OpenClaw. Then you’ll install Hermes, configure it with your chosen local model, and run a complete agent session from your terminal including tool calls, persistent memory, and auto-generated skills the agent builds from your work. By the end, you’ll have a working local AI agent on your machine, hands-on experience comparing it against the main alternatives, and a clear sense of which tool to reach for next time you want to keep AI work on-device.

This live event is for you because...

  • You’re an AI or ML engineer who wants a guided first build with a modern framework.
  • You’re a developer who wants privacy-preserving, low-cost AI agents running on your own hardware for product features or internal tools.
  • You’re an indie developer or technical founder interested in AI agents that run on a small VPS or local machine without recurring API fees.

Prerequisites

  • Python 3.11+ with the uv package manager installed on your machine
  • Have Ollama installed and pull at least one model based on your hardware (recommended [32 GB+ RAM]: ollama pull qwen3.6:35b-a3b or alternates: ollama pull gemma4:31b, ollama pull llama4:maverick; small-hardware fallbacks [8–16 GB RAM]: ollama pull qwen3:8b or ollama pull gemma4:7b)
  • Hermes Agent installed
  • The course repo cloned (link shared one week before the session)
  • Basic familiarity with LLMs
  • Comfortable in a terminal
  • Experience with any AI coding assistant (helpful but not required)

Recommended follow-up:

Schedule

The time frames are only estimates and may vary according to how the class is progressing.

The local AI agent landscape (30 minutes)

  • Presentation: Why local agents now?; the practical case for running agents on your own machine; what changed in 2026; a quick mental map (local models, harnesses, and where Hermes fits in); picking your local model; a simple rubric for picking a model (RAM budget, task type, tool-calling reliability needs); the local coding-agent landscape and where each one shines
  • Q&A
  • Break

Your first Hermes Agent (30 minutes)

  • Hands-on exercises: Install Hermes and connect it to your local model; pip install hermes-agent and point it at Ollama; run hermes model to pick your local model; write a minimal SOUL.md (agent personality) and a project context file; enable filesystem and ripgrep tools via Hermes tools and run a coding/exploration task end-to-end; watch tool calls land against your local model and see how Hermes’s per-model parsers handle different model families’ tool-call formats
  • Q&A
  • Break

Memory and skills in action (30 minutes)

  • Presentation: What Hermes uniquely brings; FTS5 session search across all prior conversations; agent-curated long-term memory and the curator process that autogenerates and refines reusable skills; where memory lives
  • Hands-on exercise: Make the agent learn; tour the agentskills.io open standard and the community Skills Hub
  • Q&A

MCP, deployment, and tool comparison (25 minutes)

  • Presentation: Extending Hermes; wiring an MCP server (filesystem, web fetch) into Hermes; when to use local, Docker, Modal, or Daytona; tasks that survive crashes and reboots; the gateway to make the same agent reachable from CLI, Telegram, Slack, Discord, and email
  • Demonstration: Deploying Hermes to a $5 VPS with Modal hibernation; topology walkthrough and how serverless backends bring idle cost near $0
  • Q&A
  • Break

When to use Hermes, Claude Code, and OpenClaw (5 minutes)

  • Presentation: A decision matrix for when Hermes is best (persistent workflows, learning from work, multiplatform access), when Claude Code plus Ollama is best (deep IDE integration, mature tooling), and when OpenClaw is best (lightest setup, pure CLI coding); pointers to the official docs, the awesome-hermes-agent community list; the Hermes self-evolution repo
  • Q&A

Your Instructor

  • Lucas Soares

    Lucas Soares is a machine learning engineer who has worked at K1 Digital and Biometrid, where he developed computer vision and NLP models for applications such as document verification, OCR-based applications, and recommender systems. Lucas has also developed various ML models, including neural networks, Siamese networks, convolutional neural networks, LSTMs, and genetic algorithms.

    linkedinXlinksearch

Skill covered

Cloud Native