Using Notebooks (Jupyter & Colab)¶
Hands-On AI is just Python, so it works anywhere Python does, notebooks included. There's only one thing to think about: where the model runs.
Want to dive straight in? Open the starter notebook:
It walks through chat, personality bots, memory, a tool-using agent, and a workflow, in a few short cells. The rest of this page explains the setup it uses.
For a deeper look at retrieval, the
RAG notebook
builds a search index over your own documents and grounds a model's answers in
them (it also needs an embedding model, e.g. ollama pull nomic-embed-text).
!pip install hands-on-ai
from hands_on_ai.chat import get_response
print(get_response("Explain photosynthesis like I'm 10."))
Local notebooks (Jupyter, JupyterLab, VS Code)¶
If the notebook runs on your own machine, local Ollama
works out of the box: it listens on http://localhost:11434 and Hands-On AI
connects automatically, no configuration needed. You can also point at a remote
provider if you prefer (see Choose a Provider).
Google Colab and other hosted notebooks¶
The catch with Colab (or any hosted notebook) is that the code runs on a remote
server, not your laptop, so it cannot reach an Ollama running on your own
localhost. You need a provider the notebook can reach over the internet.
The recommended classroom setup is an Ollama server with a bearer key, hosted by the educator, so a whole class can share it without anyone needing a paid API key. Put this in the first cell:
!pip install -q hands-on-ai
import os, getpass
os.environ["HANDS_ON_AI_SERVER"] = "https://ollama.your-school.edu"
os.environ["HANDS_ON_AI_API_KEY"] = getpass.getpass("Class API key: ")
os.environ["HANDS_ON_AI_MODEL"] = "llama3"
from hands_on_ai.chat import get_response
print(get_response("Hello from Colab!"))
A few notes:
- Set the environment variables before your first
get_responsecall. A config cell at the top of the notebook is the simplest place. - Use
getpass.getpass(...)for the API key rather thaninput(...), so the key stays out of the saved notebook and its output. - Any OpenAI-compatible provider works the same way (OpenAI, Groq, OpenRouter, and so on). Just swap the server URL, key, and model. See Choose a Provider.
- The one
pip install hands-on-aicovers everything, including RAG's PDF and Word support. There are no extra installs to remember.
For educators¶
Hosting one Ollama server with a bearer key is usually the smoothest classroom path: students only paste a URL and a key, and no one needs their own cloud account. The Choose a Provider and Classroom Setup guides cover the details.