TogetherEmbeddings: Reference Guide

What is TogetherEmbeddings?

TogetherEmbeddings connects LangChain to Together AI's embedding service, which hosts a variety of open-source embedding models — including M2-BERT, BGE, and UAE-Large-V1 — through a single API. The class wraps Together AI's /embeddings endpoint and follows the standard LangChain Embeddings interface, making it a drop-in replacement anywhere OpenAIEmbeddings or HuggingFaceEmbeddings are used. The default model is togethercomputer/m2-bert-80M-32k, producing 768-dimensional vectors suited for long-context retrieval.

Together AI's pricing is significantly cheaper than OpenAI's text-embedding-3-small at scale, making TogetherEmbeddings attractive for batch indexing jobs. The class supports both embed_query() for single strings and embed_documents() for lists, mapping to single and batch API calls respectively. Together AI's batch API processes up to 256 texts per request; LangChain handles chunking automatically for larger lists via the chunk_size parameter.

The main tradeoff is model quality. Together AI's hosted open-source models perform well on general retrieval benchmarks but trail OpenAI's text-embedding-3-large on domain-specific tasks. For RAG pipelines where query-document relevance is critical — legal, biomedical, financial — benchmark on your own dataset before committing. Set TOGETHER_API_KEY as an environment variable; the class reads it via os.getenv and raises ValueError at construction time if it is absent.

When to Use

You want affordable embeddings with good quality. Use for budget-conscious projects.

Use Cases

• • Cost-effective RAG
• • Budget embeddings
• • Multiple models
• • Batch processing
• • Startup projects
• • Demo applications

Key Features

• ✓ Affordable
• ✓ Multiple models
• ✓ Batch API
• ✓ Fast
• ✓ Quality
• ✓ Simple

When NOT to Use

Notes

Import

from langchain_together import TogetherEmbeddings

Configuration

Usage Examples

embeddings = TogetherEmbeddings()
vec = embeddings.embed_query('What is LangChain?')
print(f'Vector dim: {len(vec)}')

from langchain_together import TogetherEmbeddings

embeddings = TogetherEmbeddings(
    model='togethercomputer/m2-bert-80M-32k-retrieval',
    chunk_size=256,
)
docs = [
    'LangChain is a framework for LLM apps.',
    'Together AI hosts open-source models.',
    'RAG pipelines retrieve relevant context.',
]
vectors = embeddings.embed_documents(docs)
print(f'Embedded {len(vectors)} docs, dim={len(vectors[0])}')

from langchain_together import TogetherEmbeddings
from langchain_chroma import Chroma

embeddings = TogetherEmbeddings(model='UAE-Large-V1')

store = Chroma.from_documents(
    documents=docs,
    embedding=embeddings,
    persist_directory='./chroma_together',
)
results = store.similarity_search('affordable LLM hosting', k=3)
for r in results:
    print(r.page_content)

Common Pitfalls

Alternative Embedding Models

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 TogetherEmbeddings and the wider framework.

TogetherEmbeddings FAQ