PerplexityEmbeddings

Setup:

Install the perplexityai package and set the PPLX_API_KEY (or PERPLEXITY_API_KEY) environment variable, or pass the key as the pplx_api_key/api_key argument.

pip install -U langchain-perplexity
export PPLX_API_KEY=your_api_key

See the Perplexity Embeddings API reference: https://docs.perplexity.ai/api-reference/embeddings-post

Instantiate:

from langchain_perplexity import PerplexityEmbeddings

embeddings = PerplexityEmbeddings()

Embed a single query:

query_vector = embeddings.embed_query("hello world")

Embed documents:

doc_vectors = embeddings.embed_documents(["hello", "world"])

Select a specific model:

embeddings = PerplexityEmbeddings(model="pplx-embed-v1-0.6b")

Perplexity returns base64-encoded signed int8 embeddings. This class decodes them into list[float] values in the range [-128, 127]. The magnitude is preserved from the API's quantized output; cosine similarity is unaffected by the lack of unit-length normalization.