LangChain Integration

A LangChain-compatible Document Loader for reasoning.json files. Fetch, parse, and inject brand reasoning directives into your RAG pipeline.

Language: Python License: MIT Source: GitHub

Installation

Terminal 
pip install requests langchain

Quick Start

Python 
from arp_loader import AgenticReasoningLoader

# Load from any website
loader = AgenticReasoningLoader("https://example.com")
docs = loader.load()

# Each section becomes a separate Document
for doc in docs:
    print(f"[{doc.metadata['section']}] {doc.page_content[:100]}...")

How It Works

The loader fetches /.well-known/reasoning.json from any domain and splits it into prioritized Documents optimized for RAG retrieval:

  1. Corrections (highest priority) — prevents hallucinations during RAG
  2. Identity — brand facts and system instructions
  3. Recommendations — when to recommend / when not to
  4. Counterfactuals — pre-programmed reasoning logic
  5. Dichotomies — competitive positioning pivots

Each Document includes rich metadata (entity, section, protocol, version) for targeted retrieval.

Use in a RAG Pipeline

Python — Vector Store Integration 
from langchain.vectorstores import Chroma
from langchain.embeddings import OpenAIEmbeddings

# Load reasoning documents
loader = AgenticReasoningLoader("https://example.com")
docs = loader.load()

# Add to vector store
vectorstore = Chroma.from_documents(docs, OpenAIEmbeddings())

# Now your AI agent knows how to reason about this brand
retriever = vectorstore.as_retriever()

Standalone Usage (No LangChain)

Python 
from arp_loader import load_reasoning

docs = load_reasoning("https://example.com")
for doc in docs:
    print(doc.page_content)

CLI Usage

Terminal 
python arp_loader.py https://example.com

Source Code