ppx-langchain

LangChain integration for the Preference Profile Exchange (PPX).

Install

pip install ppx-langchain

Requires Python 3.11+. Published version: 0.1.0a1 (PyPI).

Alpha, tracking a draft specification. Expect breaking changes. The version is 0.1.0a1 on PyPI and 0.1.0-alpha.1 on npm — the same release in each ecosystem's required format.

Pulls in ppx-client and langchain-core. No LangChain models are required — bring your own (OpenAI, Anthropic, Ollama, etc.).

Three ways to use PPX in a LangChain app

1. As a document loader — load claims once

from ppx_client import PpxClient
from ppx_langchain import PpxClaimLoader

client = PpxClient("https://api.provider.app")
loader = PpxClaimLoader(client, user_token=my_keycloak_token)
docs = loader.load()  # one Document per claim

2. As a retriever — query claims per question

from ppx_langchain import PpxRetriever

retriever = PpxRetriever(
    client=client,
    grant_token=my_grant_token,
    default_context={"climate": "hot_humid"},
    requested_namespaces=["core", "fragrance"],
)

docs = retriever.invoke("What scent profile suits this user?")

3. As a tool — let the agent call it

from langchain_openai import ChatOpenAI
from ppx_langchain import ppx_preference_tool

llm = ChatOpenAI(model="gpt-4o-mini")
tool = ppx_preference_tool(client, grant_token=my_grant_token)
agent = llm.bind_tools([tool])

response = agent.invoke(
    "Recommend a fragrance. Use lookup_user_preference to check core + fragrance traits."
)

Every read is grant-scoped. The agent can never see claims the user's grant doesn't permit.

License

Apache-2.0.