Embeddings API

The embeddings API creates embeddings of one or more strings. Pico AI Server supports both OpenAI and Ollama compatible endpoints.

The embeddings API is available from Pico 1.1.12 and newer

Supported models

Unlike LLMs, embedding models are downloaded and loaded into memory dynamically when required. Pre-downloading and persistent caching of embedding models is neither necessary nor supported in Pico AI Server. Due to their compact size, embedding models can be efficiently retrieved on demand.

By default, embedding models are stored in <base-directory>/Embeddings as specified in the Models section of Settings.

Embedding models leverage the CoreML-based Swift-Embeddings library. For more implementation details, refer to the project’s GitHub repository.

Name	Architecture
all-MiniLM-L6-v2	BERT
msmarco-bert-base-dot-v5	BERT
LaBSE	BERT
gte-base	BERT
bert-base-uncased	BERT
mxbai-embed-large-v1	BERT
roberta-base	RoBERTa
xlm-roberta-base	XLM-RoBERTa
multilingual-e5-large	XLM-RoBERTa
multilingual-e5-small	XLM-RoBERTa
paraphrase-multilingual-mpnet-base-v2	XLM-RoBERTa
xlm-roberta-base-multilingual-en-ar-fr-de-es-tr-it	XLM-RoBERTa
clip-vit-base-patch16	CLIP
clip-vit-base-patch32	CLIP
clip-vit-large-patch14	CLIP
glove-twitter-25	Word2Vec
glove-twitter-50	Word2Vec
glove-twitter-100	Word2Vec
glove-twitter-200	Word2Vec
potion-base-2M	Model2Vec
potion-base-4M	Model2Vec
potion-base-8M	Model2Vec
potion-retrieval-32M	Model2Vec
potion-base-32M	Model2Vec
M2V_base_output	Model2Vec
M2V_base_output	Static
static-similarity-mrl-multilingual-v1	Static

OpenAI-compatible endpoint

Request

POST /v1/embeddings

Name	Type	Description
model	String	Name of the embeddings model, e.g. all-MiniLM-L6-v2
input	String or array of strings	String or strings to embed
encoding_format	Optional string	This property is ignored
user	Optional string	This property is ignored

Response

Name	Type	Description
model	String	Name of the embeddings model, e.g. all-MiniLM-L6-v2
object	String	Always list
data	Array of embedding objects	See embedding objects

Name	Type	Description
index	Integer	Index of the embedding
object	String	Always embedding
embedding	Array of floats	Embedding

Ollama-compatible endpoint

Request

POST /api/embed

Name	Type	Description
model	String	Name of the embeddings model, e.g. all-MiniLM-L6-v2
input	String or array of strings	String or strings to embed

Response

Name	Type	Description
model	String	The model used to generate the embeddings
embeddings	Array of array of floats	The generated embeddings
total_duration	Integer	Always nil
load_duration	Integer	Always nil
prompt_eval_count	Integer	Always nil