Setup

Ten minutes. No API key needed for anything in Parts 1 through 3.

1. Lab code

git clone https://github.com/brianbaldock/graph-engineering-course
cd graph-engineering-course/labs
python3 -m venv .venv
.venv/bin/pip install -r requirements.txt

2. Confirm it works

.venv/bin/python -m pytest tests/ -q

Expected:

............................                                             [100%]
50 passed in 0.10s

3. Run the pipeline

.venv/bin/python -m graphlab.pipeline

You should see episodes ingested through the validation gate, a temporal close, and three retrieval queries with cited edges.

4. Pick a driver

You only need one.

GitHub Copilot CLI

Model-agnostic, one subscription. Best for feeling the routing idea.

copilot --version
gh auth status

Hermes Agent

Native MCP client plus durable memory. Best for ambient graph access in every conversation.

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

Hermes talks to the lab over MCP. If its optional mcp dependency is missing, MCP is silently disabled rather than erroring, so confirm the tools appear before blaming your config. Lesson 1 has the full walkthrough.

Optional: production track

Only needed for Lesson 10. Docker for Neo4j, uv for the Graphiti MCP server, and a provider API key if you want real LLM extraction.

Use absolute paths. MCP servers launch as subprocesses with a working directory you do not control. Relative paths in an MCP config are the most common setup failure in both drivers.

Start with Lesson 0 →