Afterschool Agent · Concepts

The words, made plain.

Agent, chatbot, harness, LLM, skills, profiles, tool calls, and the cloud-versus-local choice — explained simply.

Narration
A three minute plain-words narration.
The core promise
“No jargon. Just what each piece is for, and how to use it.”
Setup requirements

What you need before you build.

Part four
Accounts & servicesThings you sign up for.
here.now — free account
The publishing house. Where the agent uploads the pages and audio it builds, so you get a web address you can open. No cost to start.
Jina.ai — free, no card
Gives the agent its web search. When it needs to look something up, it uses Jina. The free tier is plenty for a young person’s daily work.
OpenRouter.ai — ~$20 credit
The cloud brain. Your credit funds the agent’s thinking with a cloud model, and covers the paid extras like turning a lesson into a song or short video. It draws down slowly, not all at once.
LM Studio — local brain
Three models that run on a machine with eight gigabytes of RAM. How the agent runs entirely on your device, no cloud, no monthly cost. If your computer cannot handle it, use OpenRouter instead.
Piper TTS — the voice
Free, local text to speech. Reads the agent’s words aloud each morning. Runs on the device, no cloud, no account.
Telegram or WhatsApp — delivery channel
The Morning Minute and Check-In have to reach you somewhere. Set up a Telegram bot, or a WhatsApp bridge. This is how the daily rituals arrive.
Software on the machineWhat you install.
Hermes Agent — the core
The most important piece. The harness, the frame the whole agent runs on. Installing it is step one of the entire build. Without it, there is no agent.
Python — runtime
A working install. Hermes depends on it. On Windows you check it with python, dash dash, version.
Memory provider — e.g. Mnemosyne
Where the agent stores what it learns about you. The recommended one, Mnemosyne, is local, private, and needs no account. Your facts stay on your device.
A capable computer
Match the requirements to the hardware. The all-local path needs enough memory for a local model. On a modest machine, lean on OpenRouter instead.
The orderSign up for here.now, Jina and OpenRouter, set up LM Studio and Piper. Install Hermes and Python. Pick a delivery channel. Choose a memory provider. Then you are ready to build.
The big picture

Agent, chatbot, harness, LLM, or … where it all fits.

Here is the whole map in one breath. The agent is a worker that plans and acts. The harness is the frame it runs inside, holding its tools and memory. The LLM is the brain that understands words. The brain runs either on your machine or in the cloud. Everything else — skills, the user profile, the personality, tool calls — is how you steer and shape the worker. Scroll down for each term in plain words.

The core ideas

What each thing is, plainly.

AI Agent — a worker with a desk and a to-do list
Give it a job. It plans how to do it, uses tools to get it done, and checks back with you. It does not just talk, it acts.
Chatbot vs Agent — talking vs working
A chatbot talks. You ask, it answers, that is it. An agent works. It looks things up, reads files, runs tasks, keeps memory, and carries a job through. The chatbot is a receptionist. The agent is a colleague.
Harness — the frame the agent runs inside
The workspace that holds the tools, the memory, the schedule, and the safety rails. The agent is the worker, the harness is the building with the tools in it. A Hermes profile is the harness tuned for one person.
LLM — the brain, large language model
The part that understands words and writes back. Trained on a huge amount of text. It can read, reason, and produce language. But it only thinks. It does not act on its own. That is the agent’s job, using the brain.
Local LLM — runs on your machine
Runs offline, on your own device. Your data never leaves home. Private, and free to run, but it needs a capable computer.
Cloud-based LLM — runs on a provider’s servers
Used over the internet. More powerful, runs on nearly any device, but your words cross the network and it usually costs a little per use. Services like OpenRouter, DeepSeek, and OpenAI offer these.
Skills.md File — a playbook
A set of instructions the agent reads when it needs a certain kind of job. Write a skill and you teach the agent a new, clear set of steps.
User Profile (dot md) — the file that remembers you
Holds the facts about you so the agent remembers what you like, who you are, and what matters to you.
Agent Personality — the character it speaks with
The tone, the warmth, the way it talks. Same brain, different voice.
Tool Call — the agent reaching for a tool
What happens when the agent decides to do something, not just say it. It asks the harness to run a tool, search the web, open a file, send a message, takes the result, and carries on.
Talking to the world

API, webhook, API key, web fetch, web search, and memory.

Part two
API — a doorway between programs
A set of clear rules that lets one piece of software ask another for something and get the answer back. Like a waiter in a restaurant: you read the menu, tell the waiter what you want, and it brings your order from the kitchen without you going in yourself.
Webhook — being told when something happens
The opposite of asking. Instead of you checking over and over, the other program pings you the moment something changes. Someone mentions you online, a payment goes through, a file uploads, the webhook tells your agent and it can act. Polling is asking again and again; a webhook is being told once.
API Key — your secret passcode
A long code proving you are allowed to use an API. When you sign up with a provider, they give you one, and you include it with your requests so they know it is really you and can count your usage. Guard it like a password. If someone else gets it, they can use your account.
Web Fetch — downloading one known page
The agent goes to a specific address and reads that page. Best when you know exactly which page you want. Fetch is going to one address.
Web Search — finding the pages you need
The agent takes a question, finds relevant pages across the web, and reads the best ones. Search is finding the address in the first place.
Memory — how the agent keeps what it learns
The agent stores facts about you, your name, interests, goals, what you like and do not like, so the next time it talks to you it does not start from nothing. Memory can live on your device (local, private, nothing leaves home — what the Afterschool Agent uses) or in a cloud store.
Using the brains

Cloud LLMs and local LLMs, how to actually use them.

CLOUD

Use a cloud-based LLM

Connect the agent to a provider’s key, choose a model, and it calls that brain over the internet when it needs to think. Add the key, pick the model, test it. Powerful and works on any device.

LOCAL

Use a local / edge LLM

Download a model to your machine and point the agent at it. Runs offline, no key, no per-use cost, and your data stays home. Download the model, point the agent at it, test it. More privacy, more computer power needed.

The trade-off, plainlyCloud: stronger brain, easier, but your words cross the network and it can cost. Local: private and free, but needs a beefy machine. Both produce an agent that genuinely helps the people it was built for. The Afterschool Agent defaults to local-first so the young person’s data never leaves their device.
Artifacts

The real things the agent builds for you.

Part three

An artifact is a finished piece of work the agent produces — not just an answer, but something real you can open, listen to, watch, and keep. The agent does not only talk. It builds.

HTML

here.now Pages

The agent’s publishing house. It uploads a page to the here.now drive and publishes it to a web address. Beautiful learning pages, guides, hubs, index pages. This site was made this way.

AUDIO

Spoken Files

The agent writes a script and speaks it with a text-to-speech voice — like the narration on this page. Morning minutes, prayers, briefings, guides you can listen to rather than read.

VIDEO

Motion

The agent can take a page, move and animate it, and render it into a short video. Explainer clips and motion versions when showing is easier than telling.

DOCS

Documents

Word documents, spreadsheets, and PDFs. Plans, notes, worksheets, checklists, guides — things you can download, edit, print, or share.

In the workflowsEach Afterschool Agent ritual makes something real. The Morning Minute makes an audio briefing. The planner makes a document you can open and change. A lesson can become a here.now learning page, a video, or a short song. The agent picks the artifact that fits the job, makes it, and hands you a link or a file to keep. The harness is the workshop, the brain is the thinker, and the artifacts are what come out the door.
Put it together

One worker, all its parts.

Back to the courseThis is the vocabulary the build path uses. When a module says “install the harness” or “choose your brain,” you now know what those mean. Return to the overview →