examples

QueryPilot examples

Runnable, self-contained examples covering QueryPilot's core surface. Every example uses the bundled demo SQLite fixture (tests/fixtures/demo.db), so none of them need a real production database.

Setup

# From the repo root
python3 -m venv .venv
.venv/bin/pip install -e ".[dev,eval]"

# Seed the demo fixture if it is missing
.venv/bin/python tests/fixtures/seed_demo.py

Then run any example from the repo root, e.g.:

.venv/bin/python examples/01_quickstart.py

Index

Example	What it shows	How to run	Needs a key?
01_quickstart.py	Connect to the demo fixture, execute_sql, offline ask(), print validation risk level	python examples/01_quickstart.py	No
02_openai_tool_use.py	as_openai_tools() wired into an OpenAI (gpt-5.1) tool-use loop	pip install -e ".[openai]" then python examples/02_openai_tool_use.py	OPENAI_API_KEY
03_anthropic_tool_use.py	as_anthropic_tools() wired into an Anthropic (claude-sonnet-5) tool-use loop	pip install -e ".[anthropic]" then python examples/03_anthropic_tool_use.py	ANTHROPIC_API_KEY
04_access_control.py	AccessPolicy: blocked columns, row filter, masking — a rejected query and a masked result	python examples/04_access_control.py	No
05_custom_eval_suite/	A custom YAML suite (3 correctness + 1 safety case) run with querypilot eval run / check	bash examples/05_custom_eval_suite/run.sh	No
06_mcp/	Run querypilot mcp and a paste-ready Claude Desktop / Claude Code MCP config	see 06_mcp/README.md	No
07_frameworks.py	The same safe tools as LangChain, OpenAI Agents SDK and Claude Agent SDK tools	pip install -e ".[langchain,openai-agents,claude-agent]" then python examples/07_frameworks.py	No

Examples 02 and 03 build the QueryPilot connection and tool schemas before their API-key guard, so you can confirm the schema wiring even without a key — they exit with a clear message when the key is absent.

_common.py is a tiny shared helper that resolves the demo fixture path; it is not part of QueryPilot's public API.