LangChain loader (Python)

pip install spdf-langchain

pip install -e python/ -e "integrations/langchain-python[test]"

from spdf_langchain import SpdfLoader

loader = SpdfLoader("spdf-in-five-pages.spdf", locale="en")   # a file, a folder or a list of both
docs = loader.load()                  # or: for doc in loader.lazy_load(): ...

docs[0].page_content                  # the literal passage, exactly as in the source
docs[0].metadata["citation"]          # '(Saorín Ferrer, 2026, p. 1)'
docs[0].metadata["anchor_uri"]        # 'spdf:sha256-50d9…5f4c#p=2&f=1&char=15,307'
docs[0].id                            # 'sha256-50d9…5f4c:f2-1' (stable across runs)

import hashlib
import math
import re

from langchain_core.embeddings import Embeddings
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.vectorstores import InMemoryVectorStore

from spdf_langchain import SpdfLoader


class HashingEmbeddings(Embeddings):
    """A toy bag-of-words embedding: no model to download, no API key."""

    def __init__(self, dim: int = 512) -> None:
        self.dim = dim

    def _vec(self, text: str) -> list[float]:
        v = [0.0] * self.dim
        for word in re.findall(r"\w+", text.lower()):
            v[int(hashlib.md5(word.encode()).hexdigest(), 16) % self.dim] += 1.0
        norm = math.sqrt(sum(x * x for x in v)) or 1.0
        return [x / norm for x in v]

    def embed_documents(self, texts: list[str]) -> list[list[float]]:
        return [self._vec(t) for t in texts]

    def embed_query(self, text: str) -> list[float]:
        return self._vec(text)


docs = SpdfLoader("integrations/fixtures/spdf-in-five-pages.spdf", locale="en").load()
store = InMemoryVectorStore.from_documents(docs, embedding=HashingEmbeddings())  # needs numpy

question = "How is a plate without a printed folio cited?"
hits = store.similarity_search(question, k=2)
for d in hits:
    print(d.metadata["citation"], d.metadata["anchor_uri"])

prompt = ChatPromptTemplate.from_messages([
    ("system", "Answer from the passages only. After each claim, copy the citation of its passage."),
    ("human", "{context}\n\nQuestion: {question}"),
])
context = "\n\n".join(f"{d.page_content} {d.metadata['citation']}" for d in hits)
messages = prompt.invoke({"context": context, "question": question})
# answer = chat_model.invoke(messages)

(Saorín Ferrer, 2026, p. [3]) spdf:sha256-50d94244…5f4c#p=4&f=3&char=0,133
(Saorín Ferrer, 2026, p. 2) spdf:sha256-50d94244…5f4c#p=3&f=2&char=19,258

Plate I. A page with its folio and a manicule pointing at a passage. This plate carries no printed number; its folio, 3, is inferred. (Saorín Ferrer, 2026, p. [3])

Every unit records who read it: … the citation puts it in brackets. (Saorín Ferrer, 2026, p. 2)

from langchain_community.vectorstores import FAISS          # pip install langchain-community faiss-cpu
from langchain_huggingface import HuggingFaceEmbeddings      # pip install langchain-huggingface

from spdf_langchain import SpdfLoader

docs = SpdfLoader("library/", with_vectors="all-MiniLM-L6-v2@384").load()
pairs = [(d.page_content, d.metadata.pop("vector")) for d in docs]   # keep the metadata flat
store = FAISS.from_embeddings(
    pairs,
    HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2"),  # embeds queries only
    metadatas=[d.metadata for d in docs],
    ids=[d.id for d in docs],
)
print(store.similarity_search("inferred folio", k=1)[0].metadata["citation"])

cd integrations/langchain-python
uv venv && uv pip install -e ../../python -e ".[test]"
.venv/bin/python -m pytest