Building community agents with NanoClaw

Course overview

This course introduces:

1. What a basic development setup looks like, and what each tool in it is for.
2. The parts of an AI agent: the model, the harness, tool calls and data, and the orchestrator that runs them.
3. A data source: Notion, as the place an agent keeps what it learns.
4. Customising an agent by asking in plain words, and customising NanoClaw itself with skills.
5. Rules for an agent: who may ask it for what, and the difference between rules an agent follows and rules the system enforces.

By the end of this course, you should be able to:

1. Know your tools. Describe the software you install to build a community agent, and what each piece is for.
2. Codex and ChatGPT. Explain how they are related: your ChatGPT plan pays for OpenAI's model, and Codex is the program that puts it to work.
3. Install NanoClaw. Set it up with Telegram as the chat app and Codex as the harness, and make an agent from the community-assistant template.
4. Connect a data source. Create a Notion connection, and give an agent a Notion page to read and write.
5. Run an agent in a group. Put an agent in a group chat, and see how it treats a new member, a verified neighbour and an admin differently.
6. Customise in plain words. Change an agent's personality, and schedule a message, just by asking.
7. See how it works inside. Find where those changes are stored on your computer, and describe how this course's NanoClaw was customised with skills.
8. Govern the data, and go further. Explain the difference between data governance and data control, and the options for running an agent somewhere other than your laptop.

Lessons

1. Getting started: get your computer ready, with one script instead of one page per tool:
   1. Install WSL (Windows pre-requisite)
   2. Run the installation script
   3. Sign in to Codex
2. Setting up NanoClaw: create a Telegram bot, install NanoClaw, and say hi to your first agent. One page per step:
   1. Create your Telegram bot
   2. Download NanoClaw
   3. Run NanoClaw's setup
3. Setting up data sources: give Louis, your agent, a place to keep what they learn, using a ready-made Notion template. One page per step:
   1. Create the Notion page
   2. Create a Notion connection
   3. Connect your Notion page
   4. Connect your agent to Notion
4. Customising your agent: put Louis in front of real people, and make them your own. One page per step:
   1. Create a Telegram group
   2. Add your agent to the group
   3. Add humans to the group
   4. Update your agent's personality
   5. Schedule a message
5. Concept recap: step back and see what you built, how it fits together, and where to go next. One page per topic:
   1. What we did
   2. How it fits together
   3. Inside NanoClaw
   4. Our modified NanoClaw
   5. Where to go next
6. Cleaning up: remove what the course put on your computer, when you no longer need it.
   1. Uninstalling NanoClaw
   2. Uninstalling the tools

References

Title	Description	URL
NanoClaw GitHub	GitHub repository for NanoClaw	https://github.com/nanocoai/nanoclaw
NanoClaw for this course	Our fork of NanoClaw, changed to work with Telegram and Codex	https://github.com/aijutsu/aith-nanoclaw-codex-telegram
NanoClaw Website	Website for NanoClaw	https://nanoclaw.dev/
Install WSL	Microsoft's guide to installing WSL on Windows	https://learn.microsoft.com/en-us/windows/wsl/install
Install Git	Git's download page, with steps for every system	https://git-scm.com/install/
Homebrew	The app store for the terminal on a Mac	https://brew.sh/
OrbStack	A lighter way to run Docker on a Mac	https://docs.orbstack.dev/
Docker Desktop	Docker's guide to installing Docker Desktop	https://docs.docker.com/desktop/
Docker Engine on Ubuntu	Docker's guide to installing Docker on Ubuntu	https://docs.docker.com/engine/install/ubuntu/
GNU Make	The manual for Make, the tool that runs a long command from a short name	https://www.gnu.org/software/make/manual/
jq	The tool NanoClaw's setup uses to read the answers that services send back	https://jqlang.org/
Codex CLI	OpenAI's guide to installing and using Codex in the terminal	https://learn.chatgpt.com/docs/codex/cli
Claude Code	Anthropic's guide to installing Claude Code	https://code.claude.com/docs/en/setup
Telegram bots	Telegram's guide to bots and BotFather	https://core.telegram.org/bots/features#botfather
BotFather	Telegram's official bot for making bots, opened in Telegram	https://t.me/BotFather