Installation¶

agentprdiff is published on PyPI and installs cleanly on any Python 3.10+ environment with no compiled dependencies.

Prerequisites¶

Requirement	Version	Notes
Python	3.10, 3.11, or 3.12	Tested on each in CI.
`pip`	≥ 23.0	For modern resolver behavior.
Git	any	Baselines live in your repo, so you'll commit them.
`OpenAI` SDK	optional, ≥ 1.0	Only if using the OpenAI adapter or `openai_judge`.
`Anthropic` SDK	optional, ≥ 0.30	Only if using the Anthropic adapter or `anthropic_judge`.

The package itself depends on click, rich, pydantic (v2), and pyyaml — all pure-Python.

Install from PyPI¶

pip install agentprdiff

Verify:

agentprdiff --version
# agentprdiff, version 0.2.3

Optional extras¶

# OpenAI / OpenAI-compatible providers (Groq, Gemini, OpenRouter, Ollama, ...)
pip install "agentprdiff[openai]"

# Anthropic Messages API
pip install "agentprdiff[anthropic]"

# Both (for a polyglot agent or to use both judge backends)
pip install "agentprdiff[openai,anthropic]"

The base wheel imports openai / anthropic lazily, so you only pay the import cost when you actually call an SDK adapter or judge.

Install from source (contributors)¶

git clone https://github.com/vnageshwaran-de/agentprdiff
cd agentprdiff
pip install -e ".[dev]"

The dev extra brings in pytest, pytest-cov, ruff, and mypy.

Run the bundled smoke test¶

cd examples/quickstart
agentprdiff init
agentprdiff record suite.py
agentprdiff check  suite.py

If the last command exits 0, your install is healthy. The example agent is fully self-contained — no API keys required.

Environment variables¶

agentprdiff itself reads only one env var. Your agent and the semantic judge read a few others.

Variable	Read by	Purpose
`AGENTGUARD_JUDGE`	`agentprdiff` (semantic grader)	`fake`, `openai`, or `anthropic`. Forces the default judge backend, ignoring autodetection.
`OPENAI_API_KEY`	`openai_judge` (when selected); your agent	Real-LLM judge calls and any agent that uses the OpenAI SDK.
`ANTHROPIC_API_KEY`	`anthropic_judge` (when selected); your agent	Real-LLM judge calls and any agent that uses the Anthropic SDK.
Whatever your agent reads	your agent	`agentprdiff` does not touch your agent's keys; it just invokes the callable.

Default judge selection rules¶

When semantic(...) runs without an explicit judge= argument, the backend is chosen in this order:

AGENTGUARD_JUDGE=fake → fake_judge
AGENTGUARD_JUDGE=openai or OPENAI_API_KEY set → openai_judge()
AGENTGUARD_JUDGE=anthropic or ANTHROPIC_API_KEY set → anthropic_judge()
Otherwise → fake_judge (deterministic, free, used in CI without keys)

Run agentprdiff check with at least one semantic() grader to see a banner that prints which judge was actually selected — useful for catching the silent fake-judge fallback.

Uninstall¶

pip uninstall agentprdiff

Your .agentprdiff/ directory and any baselines committed to git remain untouched.