.embed_documents(): Reference Guide

What is .embed_documents()?

.embed_documents() is the batch embedding method on all LangChain embedding model classes. It accepts a List[str] and returns a parallel List[List[float]]—one dense vector per input text. Issuing a single batched call is far more efficient than looping over .embed_query() because it combines the inputs into the minimum number of API requests the provider allows, eliminating per-request overhead, connection setup, and often qualifying for bulk pricing.

The output dimensionality depends on the model: OpenAI text-embedding-3-small returns 1536-dimensional vectors by default (tunable via the dimensions= parameter on newer v3 models), while text-embedding-3-large returns up to 3072-d. HuggingFace sentence-transformers return whatever the underlying architecture defines—check the model card. Downstream similarity search requires that the query vector comes from the same model, same dimensions setting, and same configuration as the stored document vectors. Mixing these produces silently wrong results.

For production-scale indexing, pair .embed_documents() with CacheBackedEmbeddings to avoid re-embedding identical text on subsequent runs. The async variant .aembed_documents() fires sub-batches concurrently and can cut wall-clock time significantly when the provider supports parallel requests. Most LangChain vector store factories (Chroma.from_documents(), FAISS.from_documents()) call .embed_documents() internally, so you only call it directly when you need the raw vectors.

Use Cases

• • Index a document collection for a vector store
• • Batch encode chunks after text splitting
• • Pre-compute embeddings and cache to disk
• • Feed raw vectors to FAISS or a custom ANN index
• • Evaluate embedding quality across a corpus
• • Async pipeline embedding with aembed_documents()

Key Features

• ✓ Single batched API call for all inputs
• ✓ Far cheaper than looping embed_query()
• ✓ Returns parallel list of vectors (one per input)
• ✓ Async variant aembed_documents() for concurrent pipelines
• ✓ Used internally by all vector store factory methods
• ✓ Compatible with CacheBackedEmbeddings for disk caching

When NOT to Use

Notes

Method Signature

vectors = embeddings.embed_documents(texts)

Parameters

Return Value

[[0.012, -0.034, ...], [0.087, 0.003, ...]]

Code Examples

from langchain_openai import OpenAIEmbeddings

embeddings = OpenAIEmbeddings(model='text-embedding-3-small')

docs = [
    'LangChain is a framework for building LLM applications.',
    'RAG pipelines combine retrieval with language generation.',
    'Vector stores hold dense embeddings for similarity search.',
]

vectors = embeddings.embed_documents(docs)

print(f'Embedded {len(vectors)} docs')
print(f'Each vector has {len(vectors[0])} dimensions')

from langchain_openai import OpenAIEmbeddings
from langchain.embeddings import CacheBackedEmbeddings
from langchain.storage import LocalFileStore

store = LocalFileStore('./embed_cache')
base_embedder = OpenAIEmbeddings(model='text-embedding-3-small')
cached_embedder = CacheBackedEmbeddings.from_bytes_store(
    base_embedder, store, namespace=base_embedder.model
)

docs = ['LangChain intro text...', 'RAG explainer...']

# First call hits the API
vecs = cached_embedder.embed_documents(docs)

# Second call returns from disk cache — no API call
vecs = cached_embedder.embed_documents(docs)

import asyncio
from langchain_openai import OpenAIEmbeddings

embeddings = OpenAIEmbeddings(model='text-embedding-3-small')

async def embed_batch(texts):
    return await embeddings.aembed_documents(texts)

docs = ['doc one', 'doc two', 'doc three']
vectors = asyncio.run(embed_batch(docs))
print(len(vectors))  # 3

Common Mistakes

Related LangChain References

Browse the full LangChain API reference index to explore more classes, methods, and decorators, or start with the LangChain introduction tutorial for end-to-end context on building with .embed_documents() and the wider framework.

.embed_documents() FAQ