ACP lesson 3 chunk_size and chunk_overlap

Short Course Q&A ACP: Agent Communication Protocol
1

Lesson 3 discusses this code:

rag_tool = RagTool(config=config,  
                   chunk_size=1200,       
                   chunk_overlap=200,     
                  )

IntelliJ flags chunk_size and chunk_overlap as unused. I did some debugging and couldn’t find any code that uses these parameters. I set them to classes that raise an error when accessed and nothing was raised. I don’t think these values are actually used at all and the course is giving incorrect information.

I think the EmbedChain is what processes the PDF file and it hard-codes chunks into 2048 bytes.

2

Most likely the underlying tools have been updated such that they no longer use those parameters, but the notebook has not been updated.

Keeping the Short Course material up to date is a losing battle. The infrastructure for this area of AI development is extremely dynamic.

3

Maybe, but the email I received said it was a new course (as of 4 days ago, June 25th, 2025).

It would help if they pinned requirements. This is their requirements.txt:

crewai
crewai_tools
acp-sdk
load_dotenv
uv
nest-asyncio
colorama
smolagents

If it was the case that the API changed, pinning the reqs would alleviate that risk.

4

I just discovered there is also a pyproject.toml file which does pin reqs:

[project]
name = "my-acp-project"
version = "0.1.0"
description = "Add your description here"
readme = "README.md"
requires-python = ">=3.11"
dependencies = [
    "acp-sdk>=0.9.1",
    "colorama>=0.4.6",
    "crewai>=0.121.0",
    "crewai-tools>=0.45.0",
    "duckduckgo-search>=8.0.2",
    "load-dotenv>=0.1.0",
    "markdownify>=1.1.0",
    "mcp>=1.9.1",
    "nest-asyncio>=1.6.0",
    "requests>=2.32.3",
    "smolagents[mcp]>=1.16.1",
]

Although I had to use

requires-python = ">=3.11,<3.13"

I still see the same unused params.

5

Thanks for your report.

I’m not a mentor for the Short Courses, so I don’t keep up on developments there as well as I should.