Package org.apache.tika.inference

Class OpenAIEmbeddingEngine

java.lang.Object

org.apache.tika.inference.OpenAIEmbeddingEngine

All Implemented Interfaces:

Closeable, AutoCloseable, Initializable, EmbeddingEngine, Engine

public class OpenAIEmbeddingEngine
extends Object
implements EmbeddingEngine, Initializable

An OpenAI-compatible embeddings endpoint (POST /v1/embeddings), named in "engines". Sends up to maxBatchSize images per request as {"image": "data:<mime>;base64,..."} inputs and reads the vectors back by index. requestParameters ride every request body verbatim, for the keys a vendor adds to the shape (Jina's task, a Matryoshka dimensions); the engine reads float vectors back, so a key that changes the response shape is refused at load. The OpenAI shape has no image input, so vendors differ: imageInput is "object" for the {"image": ...} form (Jina) or "data-uri" for a bare data URI in input (LiteLLM and other gateways).

Field Summary

Fields

Modifier and Type	Field	Description
static final String	IMAGE_INPUT_DATA_URI
static final String	IMAGE_INPUT_OBJECT
static final String	MEDIA_INPUT_NONE
static final String	MEDIA_INPUT_OBJECT

Constructor Summary

Constructors

Constructor	Description
OpenAIEmbeddingEngine()

Method Summary

Modifier and Type	Method	Description
void	close()	Releases what the engine holds (a client, a native handle); nothing by default.
List<float[]>	embedImages(List<byte[]> images, List<String> mimeTypes, ParseContext context)	One request; the vectors come back in the order of the images.
List<float[]>	embedMedia(List<MediaInput> inputs, ParseContext context)	Object form only: {"audio": data-uri} or {"video": data-uri}, as Jina names them.
List<float[]>	embedTexts(List<String> texts, ParseContext context)	One request; the vectors come back in the order of the texts.
String	getApiKey()
String	getApiKeyHeaderName()
String	getApiKeyPrefix()
String	getBaseUrl()
String	getEmbeddingsPath()
String	getImageInput()	How an image goes into input: "object" or "data-uri"; see the class note.
int	getMaxBatchSize()
int	getMaxRetries()
String	getMediaInput()	Whether audio and video may be sent, and how: "none" or "object" ({"audio": ...}, {"video": ...}).
String	getModel()
Map<String,Object>	getRequestParameters()	Extra keys for every request body, as the vendor names them; see the class note.
long	getTimeoutMillis()
void	initialize()	Called after all properties have been set to allow for validation and initialization that depends on multiple properties.
void	setApiKey(String apiKey)
void	setApiKeyHeaderName(String apiKeyHeaderName)
void	setApiKeyPrefix(String apiKeyPrefix)
void	setBaseUrl(String baseUrl)
void	setEmbeddingsPath(String embeddingsPath)
void	setImageInput(String imageInput)
void	setMaxBatchSize(int maxBatchSize)
void	setMaxRetries(int maxRetries)	Retries of a 429/502/503/504 answer, with backoff inside the parse budget; 0 fails at once.
void	setMediaInput(String mediaInput)
void	setModel(String model)
void	setRequestParameters(Map<String,Object> requestParameters)
void	setTimeoutMillis(long timeoutMillis)
boolean	supportsMedia()	Whether EmbeddingEngine.embedMedia(java.util.List<org.apache.tika.inference.MediaInput>, org.apache.tika.parser.ParseContext) is implemented; a MEDIA binding on an engine without it fails at load.

Methods inherited from class java.lang.Object

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

Field Details

MEDIA_INPUT_NONE

public static final String MEDIA_INPUT_NONE

See Also:

• Constant Field Values

MEDIA_INPUT_OBJECT

public static final String MEDIA_INPUT_OBJECT

See Also:

• Constant Field Values

IMAGE_INPUT_OBJECT

public static final String IMAGE_INPUT_OBJECT

See Also:

• Constant Field Values

IMAGE_INPUT_DATA_URI

public static final String IMAGE_INPUT_DATA_URI

See Also:

• Constant Field Values

Constructor Details

OpenAIEmbeddingEngine

public OpenAIEmbeddingEngine()

Method Details

initialize

public void initialize()
                throws TikaConfigException

Description copied from interface: Initializable

Called after all properties have been set to allow for validation and initialization that depends on multiple properties.

Specified by:

initialize in interface Initializable

Throws:

TikaConfigException - if there is a problem with the configuration

embedImages

public List<float[]> embedImages(List<byte[]> images,
 List<String> mimeTypes,
 ParseContext context)
                          throws IOException,
TikaException

Description copied from interface: EmbeddingEngine

One request; the vectors come back in the order of the images.

Specified by:

embedImages in interface EmbeddingEngine

Throws:

IOException

TikaException

supportsMedia

public boolean supportsMedia()

Description copied from interface: EmbeddingEngine

Whether EmbeddingEngine.embedMedia(java.util.List<org.apache.tika.inference.MediaInput>, org.apache.tika.parser.ParseContext) is implemented; a MEDIA binding on an engine without it fails at load.

Specified by:

supportsMedia in interface EmbeddingEngine

embedMedia

public List<float[]> embedMedia(List<MediaInput> inputs,
 ParseContext context)
                         throws IOException,
TikaException

Object form only: {"audio": data-uri} or {"video": data-uri}, as Jina names them.

Specified by:

embedMedia in interface EmbeddingEngine

Throws:

IOException

TikaException

embedTexts

public List<float[]> embedTexts(List<String> texts,
 ParseContext context)
                         throws IOException,
TikaException

Description copied from interface: EmbeddingEngine

One request; the vectors come back in the order of the texts.

Specified by:

embedTexts in interface EmbeddingEngine

Throws:

IOException

TikaException

close

public void close()
           throws IOException

Description copied from interface: Engine

Releases what the engine holds (a client, a native handle); nothing by default.

Specified by:

close in interface AutoCloseable

Specified by:

close in interface Closeable

Specified by:

close in interface Engine

Throws:

IOException

getBaseUrl

public String getBaseUrl()

setBaseUrl

public void setBaseUrl(String baseUrl)

getModel

public String getModel()

setModel

public void setModel(String model)

getApiKey

public String getApiKey()

setApiKey

public void setApiKey(String apiKey)

getTimeoutMillis

public long getTimeoutMillis()

setTimeoutMillis

public void setTimeoutMillis(long timeoutMillis)

getMaxRetries

public int getMaxRetries()

setMaxRetries

public void setMaxRetries(int maxRetries)

Retries of a 429/502/503/504 answer, with backoff inside the parse budget; 0 fails at once.

getMaxBatchSize

public int getMaxBatchSize()

Specified by:

getMaxBatchSize in interface EmbeddingEngine

setMaxBatchSize

public void setMaxBatchSize(int maxBatchSize)

getEmbeddingsPath

public String getEmbeddingsPath()

setEmbeddingsPath

public void setEmbeddingsPath(String embeddingsPath)

getApiKeyHeaderName

public String getApiKeyHeaderName()

setApiKeyHeaderName

public void setApiKeyHeaderName(String apiKeyHeaderName)

getImageInput

public String getImageInput()

How an image goes into input: "object" or "data-uri"; see the class note.

setImageInput

public void setImageInput(String imageInput)

getMediaInput

public String getMediaInput()

Whether audio and video may be sent, and how: "none" or "object" ({"audio": ...}, {"video": ...}).

setMediaInput

public void setMediaInput(String mediaInput)

getRequestParameters

public Map<String,Object> getRequestParameters()

Extra keys for every request body, as the vendor names them; see the class note.

setRequestParameters

public void setRequestParameters(Map<String,Object> requestParameters)

getApiKeyPrefix

public String getApiKeyPrefix()

setApiKeyPrefix

public void setApiKeyPrefix(String apiKeyPrefix)