docs(examples): document what the examples need in order to run

[williedoran-neo4j]

Follow-up to #595, now that the services are easy to start.

examples/README.md is a pure link index: no mention of API keys, extras, Docker, or the fact that 17 examples talk to a remote demo database. What setup knowledge exists is spread across the root README's Tests section and three per-store READMEs.

Documentation

examples/SETUP.md — what you need in order to run the examples: credentials, the services, which providers are free or have a local equivalent, and the traps that cost the most time. Running from the repo root; the three examples that need PYTHONPATH=examples/data; the two that write to a read-only demo database and so fail with Forbidden as written.

examples/.env.example — a credential template, with .env variants gitignored and the template itself kept tracked.

Every count in SETUP.md was checked against the code rather than estimated — "17 examples use the demo database", "14 call load_dotenv()", "three need PYTHONPATH", the two empty splitter files.

Not documentation — three Ollama examples change behaviour

Flagging this explicitly, because it is code in an otherwise-docs PR.

They could not be run as shipped. ollama_llm.py and ollama_embeddings.py carried a literal model_name="<model_name>" placeholder that had to be edited before the example would do anything. Documenting "first edit this file" would have been documenting a defect, so the model became a command-line argument instead:

ollama pull llama3.2
python examples/customize/llms/ollama_llm.py llama3.2   # or omit it for the default

Which model you have is a property of your machine, not of the example, which is why it is an argument rather than a new hardcoded constant. Defaults are llama3.2 for the LLMs and nomic-embed-text for the embedder, so all three now run with no arguments after a single ollama pull.

One is a real default change worth calling out: ollama_tool_calls.py was not a placeholder — it hardcoded mistral:latest. It now defaults to llama3.2, so that one ollama pull covers all three examples rather than two models being needed. Both support tool calling; the docstring now says to check the model's Ollama page before substituting one that may not.

That file's run instructions also pointed at examples/tool_calls/ollama_tool_calls.py, a path that does not exist. Corrected.

Packaging

python-dotenv and requests are now declared in the examples extra. Fourteen examples call load_dotenv() and tools_retriever_example.py calls a web API with requests, but neither was declared, so both resolved only transitively — working by accident.

Also

The Weaviate and Pinecone READMEs gave python -m a slash path, which cannot work; -m takes a dotted module path. Fixed rather than documented.

Type of Change

• New feature
• Bug fix
• Breaking change
• Documentation update
• Project configuration change

Complexity

Complexity: Low

How Has This Been Tested?

• Unit tests
• E2E tests
• Manual tests

All three Ollama examples confirmed to parse their arguments and print help. Every command quoted in SETUP.md run as written, and every numeric claim in it re-derived from the code. Green on ruff, mypy --strict and the unit suite.

Checklist

• Documentation has been updated
• Unit tests have been updated
• E2E tests have been updated
• Examples have been updated
• New files have copyright header (Markdown and .env template only)
• CLA (https://neo4j.com/developer/cla/) has been signed
• CHANGELOG.md updated if appropriate

🤖 Generated with Claude Code

[AmirLayegh]

LGTM!
Left a few comments that might be worth addressing before merging.