P1 · The 7 parts in your own words
15 minCopy the table above and fill in each row with a sentence of your own and the folder where the part lives.
Done when: all seven rows are filled in, and none of the sentences were copied from the lesson text.
Lesson 1 · 15 min
Every agent, from the simplest to the most expensive, has the same seven parts: a door, a heart, an identity, a brain, a memory, hands, and a vault. Knowing these parts helps you choose better, spend less, and avoid getting locked into one product. INTELECTO is our example: small on purpose, so you can read through it.
Start lesson
Goal: explain the path of a message and say what each of the seven parts does and where it lives in INTELECTO.
Scroll the page: each step lights up a piece.
Scroll down: each step below lights up its part of the diagram.
This is where you talk to it. In INTELECTO, that’s Telegram: one token and you’re ready, with no public internet address. The agent checks Telegram for new messages, splits replies longer than 4,000 characters, and shows “typing” while it thinks.
It receives your message, thinks, uses a tool if needed, checks the result, and thinks again until it replies with text. It has a brake: at most five tool rounds per message. After that, it has to reply.
Four simple files in the workspace folder: its soul, the rules, your profile, and the initial facts. The agent reads all four with every message. Change the text, change the assistant, without programming.
The AI model. In the cloud, OpenRouter: one key gives you access to more than a hundred models. On your computer, Ollama: no cost per message and nothing is sent outside your computer.
It saves what matters in a SQLite database on your machine, with keyword search and no API cost. With each message, it finds what’s relevant and adds it to the context.
Without tools, it only chats; with them, it gets things done. The core tools: read and write files, list folders, run commands, save to memory, and search memory. Each tool describes itself, and the model reads that description to decide whether to use it.
API keys encrypted and tied to the machine, a list of blocked commands, a 60-second timeout per command, and an audit log of everything it runs.
You write in Telegram. The loop brings together identity and memory, sends them to the model, runs the tool it requests within the vault’s limits, and sends the reply back through the same channel.
The same seven pieces appear in INTELECTO, OpenClaw, Hermes, or a Jarvis you build yourself.
Click each tab.
INTELECTO is a personal assistant for one owner. It isn’t multiuser, doesn’t depend on the cloud, doesn’t need Docker, has no web dashboard, and isn’t a black box.
The plan calls for about three thousand lines and seven dependencies: you can read it.
Can I explain what happens between my message and the response? If not, it’s a black box.
A chatbot responds and forgets. An agent thinks, acts with a tool, looks at the result, and thinks again.
Without a limit, an agent can keep spinning and spending without finishing. That’s why there’s a five-round brake.
At most 5 tool rounds per message; after that, respond with text.
Each part is defined by a contract: the channel has start, send, and stop; the provider receives the conversation and tools, then returns a response.
Want WhatsApp instead of Telegram, or Ollama instead of OpenRouter? Use a different file with the same contract.
New tool = new file using the same pattern. The registry finds it automatically, without changing the core.
The feature guide organizes everything into eight areas: identity, security, channels, memory, automation, integrations, skills, and architecture.
You don’t need everything. Scheduling and web search, for example, are in the plan for the next phase.
Loop + one brain + one channel + a basic SOUL. Memory comes right after.
The core folder tree that the agent builds in the repository example.
jarvis/ channels/ # the door: terminal and Telegram agent/ # the heart: loop.py and context.py providers/ # the brain: contract + test model memory/ # the memory: SQLite with search tools/ # the hands: files, command, memory security/ # the vault: secrets and blocklist workspace/ SOUL.md AGENTS.md USER.md MEMORY.md # the identity
| Part | What it does (in my own words) | Where it lives | |-------------|---------------------------------|----------------| | Channel | | | | Loop | | | | Identity | | | | Brain | | | | Memory | | | | Tools | | | | Vault | | |
Chatbot: responds, forgets when the conversation ends, has no identity, doesn’t act, and doesn’t know you.
Jarvis: remembers past conversations, has a name and written values, uses tools, and knows your context.
Before you build, understand what you’re going to build.
Copy the table above and fill in each row with a sentence of your own and the folder where the part lives.
Done when: all seven rows are filled in, and none of the sentences were copied from the lesson text.
Explain to someone (or record an audio message) how a message travels from Telegram to the response.
Done when: you named channel, loop, identity, memory, model, and tool in the right order, without looking.
Nei opens up INTELECTO, one part at a time.
The explanation appears after your answer.
Question 1 of 4
What makes an agent more than just a chatbot?competency
Question 2 of 4
What is the five-tool-round limit for?
Question 3 of 4
You want to switch from Telegram to WhatsApp. What changes?
Question 4 of 4
Where does the assistant’s personality live?
Choose an option. The explanation appears after your attempt.