Package com.tnsai.llm.cache

Class OllamaEmbeddingProvider

java.lang.Object

com.tnsai.llm.cache.OllamaEmbeddingProvider

All Implemented Interfaces:

EmbeddingProvider

public class OllamaEmbeddingProvider
extends Object
implements EmbeddingProvider

EmbeddingProvider backed by an Ollama server's /api/embed endpoint. Enables keyless embeddings — local by default, or Ollama Cloud — so the semantic-cache / link-suggestion tiers work without an OpenAI key.

Configuration (environment)

• OLLAMA_BASE_URL — server URL (default http://localhost:11434)
• OLLAMA_API_KEY — bearer token for Ollama Cloud (optional for local)

 var provider = new OllamaEmbeddingProvider();              // nomic-embed-text @ localhost
 var cache = SemanticCache.builder().embeddingProvider(provider).build();
 

The embedding dimension is reported as the model's known default until the first successful call, after which the real vector length is used.

Task-instruction prefixes (nomic)

nomic-embed-text is contrastively trained with a per-task instruction prefix and produces a degenerate, collapsed cosine band without one. This provider prepends the matching prefix (search_query: / search_document: / clustering: / classification:) based on the EmbeddingTask passed to embed(String, EmbeddingTask); the no-arg embed(String) defaults to EmbeddingTask.DOCUMENT. Non-nomic models are passed through unprefixed so a prefix is never applied to a model not trained on it.

Constructor Summary

Constructors

Constructor	Description
OllamaEmbeddingProvider()	Local Ollama with nomic-embed-text, honouring OLLAMA_* env vars.
OllamaEmbeddingProvider(String model)	Local Ollama with a custom embedding model.
OllamaEmbeddingProvider(String baseUrl, String model, String apiKey)

Method Summary

Modifier and Type	Method	Description
float[]	embed(String text)	Generates an embedding vector for the given text.
float[]	embed(String text, EmbeddingTask task)	Generates an embedding for text conditioned on task.
List<float[]>	embedBatch(List<String> texts)	Generates embeddings for multiple texts in batch.
List<float[]>	embedBatch(List<String> texts, EmbeddingTask task)	Batch counterpart of EmbeddingProvider.embed(String, EmbeddingTask).
int	getDimension()	Returns the dimension of the embedding vectors.
String	getModelName()	Returns the name of the embedding model.

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Details

OllamaEmbeddingProvider

public OllamaEmbeddingProvider()

Local Ollama with nomic-embed-text, honouring OLLAMA_* env vars.

OllamaEmbeddingProvider

public OllamaEmbeddingProvider(String model)

Local Ollama with a custom embedding model.

OllamaEmbeddingProvider

public OllamaEmbeddingProvider(String baseUrl,
 String model,
 String apiKey)

Parameters:

baseUrl - Ollama server URL (e.g. http://localhost:11434 or https://ollama.com); falls back to the default when null

model - embedding model name (e.g. nomic-embed-text)

apiKey - bearer token for Ollama Cloud; null/blank for local

Method Details

embed

public float[] embed(String text)

Description copied from interface: EmbeddingProvider

Generates an embedding vector for the given text.

Specified by:

embed in interface EmbeddingProvider

Parameters:

text - The text to embed

Returns:

A float array representing the embedding vector

embed

public float[] embed(String text,
 EmbeddingTask task)

Description copied from interface: EmbeddingProvider

Generates an embedding for text conditioned on task.

Instruction-tuned models (e.g. nomic-embed-text) embed the same text differently per task; supplying the right task is what separates related from unrelated text. The default ignores task — correct for providers whose model has no task convention (hash-based, OpenAI text-embedding-3).

Specified by:

embed in interface EmbeddingProvider

See Also:

• EmbeddingTask

embedBatch

public List<float[]> embedBatch(List<String> texts)

Description copied from interface: EmbeddingProvider

Generates embeddings for multiple texts in batch.

Specified by:

embedBatch in interface EmbeddingProvider

Parameters:

texts - The texts to embed

Returns:

List of embedding vectors

embedBatch

public List<float[]> embedBatch(List<String> texts,
 EmbeddingTask task)

Description copied from interface: EmbeddingProvider

Batch counterpart of EmbeddingProvider.embed(String, EmbeddingTask). The default ignores task; task-aware providers override this.

Specified by:

embedBatch in interface EmbeddingProvider

getDimension

public int getDimension()

Description copied from interface: EmbeddingProvider

Returns the dimension of the embedding vectors.

Specified by:

getDimension in interface EmbeddingProvider

getModelName

public String getModelName()

Description copied from interface: EmbeddingProvider

Returns the name of the embedding model.

Specified by:

getModelName in interface EmbeddingProvider