Building your own agentic AI assistant

This series treats a personal AI assistant the way the rest of Factnetize treats any system: as something you build, measure, and own — one step at a time. The goal is simple. Start from the ChatGPT-level familiarity you already have, and end with a real agent that runs on your own PC, uses your tools, remembers your context, and does actual work on your behalf. No hype, no vendor lock-in, no magic — just the working stack, explained.

Who this is for

It is written for one specific reader: someone who knows ChatGPT but has never run an agent. You do not need:

• to be a developer
• a server
• a GPU

If you can paste a command into a terminal when told exactly what to do, you can follow along. Each part leaves you with a more capable assistant than the last — and teaches you the pieces underneath instead of hiding them behind a polished app.

No hype, just what works

You have probably seen the videos: someone casually reveals an agentic AI that supposedly runs their whole business, inbox, and life — flawlessly, in a weekend. What you never see is the config, the real cost, the failed runs, or whether any of it survives contact with a second person trying to reproduce it. This series is the opposite of that. Everything here is meant to be achievable, reproducible, and actually working on an ordinary PC — every step shown, every gotcha named, every price a dated snapshot you can check. If a thing only works in a demo, it does not go in.

Series overview

1. What is an agentic assistant, and why build your own — The starting point: what actually makes an assistant “agentic”, how it differs from a plain chatbot, and why more people run their own on an ordinary Windows 11 PC instead of renting one.
2. Installing your first AI agent on Windows 11: Hermes Agent step by step — Desktop installer or PowerShell one-liner, model setup via Nous Portal, your first real tool call, and two Windows traps explained before they cost you time.
3. Your way around Hermes Desktop: the map for everything that follows — A visual map of the interface: where settings, providers, the model picker, capabilities, and the terminal live — plus a cheat-sheet that translates every later article into the right Desktop location.
4. What your agent actually does: four real things I use it for — Before the setup deep-dive: four real tasks — summarising a YouTube video, researching a PC I then built, fixing a broken update, and setting a reminder — so you can see what the agent does before configuring more.
5. Building the foundation: working directory, Python, and config in Hermes — Set the working directory, teach your agent which Python to use when it codes using the say-confirm-save pattern, and keep config cleanly separate from your projects.
6. Connecting more model providers to Hermes: DeepSeek, Claude, and beyond — Add DeepSeek as a cheap daily driver, reach Claude through Command Code, and learn to match the right model to each task — all in Desktop settings, no command line needed.
7. Connect Hermes to Claude Code: use Claude with no API key — If you already have Claude Code installed, Hermes can use Claude with no API key at all — it calls the CLI directly, works on plain Claude Pro, and this shows the one Windows gotcha to handle.
8. Your first coding project: build an AI tools dashboard — A hands-on first build: a single-file dashboard of all your AI tools. Sketch it cheap, polish it with a strong model, and learn the one prompt trick that unlocks genuinely fresh designs. Template included.
9. Hermes tools 101: give your AI agent its first hands — What a tool actually is, the ~86 Hermes ships with, how to switch toolsets on and off per surface, the truth about keyless web search, and a safety mindset for letting an agent act on your machine.
10. Skills: teach your Hermes agent repeatable workflows — Turn repeated multi-step jobs into procedural memory: what a SKILL.md is, why the trigger is everything, writing one yourself or letting Hermes write it, activating optional skills, and auditing community skills before you install.
11. Persistent memory: your Hermes agent’s built-in brain — Give your agent a small, disciplined memory it carries into every session: what belongs in it, the roughly 2000-character budget you can raise, and the facts-not-orders habit that keeps it useful.
12. Give your AI agent an Obsidian knowledge base — Give your agent a searchable external brain: install Obsidian, enable the note-taking skill under Capabilities, point Hermes at the vault with one config line, and grow it with the “summarise and save” habit.
13. Indexing Obsidian for an AI agent: plugin, MCP server, or build your own? — Three ways to make the agent read a growing vault fast, in one pros-and-cons matrix — plus the fact-first case for building your own indexer and a ready-to-paste prompt to do it.
14. Building a vault index so your AI agent reads Obsidian fast — The hands-on build: one Python script that maps every note with resolved wikilinks and backlinks into a single JSON file, so the agent reads the whole vault in one call instead of crawling every note.
15. Set up the Hermes Telegram gateway in minutes — Reach your agent from your phone: create a bot with BotFather, lock it to your own user ID, run the gateway as a persistent background service, and use voice memos, file transfers, and group chats.
16. Scheduling tasks in Hermes: from a reminder to real automation — Make your agent proactive: schedule jobs in plain language (a three-week reminder, a nightly index rebuild, a morning briefing), with the one rule that every scheduled prompt must be self-contained.
17. Understanding MCP: how your Hermes agent plugs into everything — Connect your agent to external systems the standard way: what MCP is (the “USB-C port for AI”), your first filesystem server, the hermes mcp commands, connecting GitHub by hand, and the allowlisting habit that keeps it safe.
18. Scrapling for Hermes: get past Cloudflare walls — Add a local stealth fetcher when a site fights back: what Scrapling does, the install the official “one command” leaves half-finished, and how to verify it actually works.
19. Model routing and cost control for your AI agent: keep a heavy agent workload affordable by reasoning on a strong model and orchestrating on a cheap one, with the prompt-cache trap that quietly wipes out your savings.

More parts are in the works. This list grows as new articles are published.

Background & concepts

These evergreen articles explain the foundations behind the series. They are not steps — you can read them in any order, skip them if you already know the topic, and return to them whenever a later article uses a term that needs unpacking.

• AI tokens and API keys explained: a plain-English guide to model access — What tokens, providers, API keys, endpoints, and gateways actually are, what the major model families look like, and what you realistically pay. Read this before connecting your first paid provider.
• Command Code as a model provider for Hermes — One API key that opens up dozens of models. How to use Command Code to escape a single provider’s usage limits and spread heavy work across many models.
• How to enable zero data retention in Hermes Agent with Command Code — Lock down strict zero data retention across your whole setup with one header — and avoid the auxiliary-call pitfall that quietly defeats a provider-only configuration.

What this series covers

• Agent vs chatbot — what “agentic” really means, and when it is worth it
• Installing your first agent on Windows 11 — no server, no GPU
• Giving it a brain — connecting a model provider that fits your budget
• Tools and MCP — letting the agent reach files, the web, and your apps
• Memory and skills — the loop that makes it improve over time
• Automation on a schedule — work that runs while you sleep
• Sub-agents — delegating parallel work to keep things fast
• Connecting it to your world — mail, calendar, files, and your own site
• Personality and soul files — making the assistant unmistakably yours

New parts are added as they’re published. Bookmark this page to follow along.

Related series

• Local AI on Windows — prefer to keep everything on your own machine with no API costs? This series covers three local setups: LM Studio + AnythingLLM, Hermes via Hyper-V, and Hermes in Docker.
• WordPress + AI — once your agent is running, this series shows how to connect it to your WordPress site via MCP so it can draft posts, set SEO fields, and manage content from a chat prompt.