Class DeterministicExtractTool

java.lang.Object

ai.redouble.nucleo.harness.AbstractJob<Extraction>

ai.redouble.nucleo.tools.AbstractTool<FileRef, Extraction>

ai.redouble.demo.extract.DeterministicExtractTool

All Implemented Interfaces:

GuardedExecution, Identifiable, Job<Extraction>, Tool<FileRef, Extraction>

public class DeterministicExtractTool
extends AbstractTool<FileRef, Extraction>

The tier that costs nothing: text the code can read without a model or a native toolkit. The file's first bytes say what it is before its extension does: an Office document (a zip of XML) is unzipped and its body parts are stripped of tags; anything unsigned is read as UTF-8 text, and comes back with textLayer false when too much of it failed to decode, which is what a binary under a text extension looks like and what the classifier then judges. Images and PDFs are not read here - they go whole to the model tier - so this tier pulls no font or graphics toolkit and compiles to a native image on any OS. A file this tool cannot open is a fault of the run, not of a model, and propagates as such.

Since:

0.1 (2026-09-14)

Author:

Andrey Santrosyan

Field Summary

Fields

Modifier and Type	Field	Description
static final Set<String>	TEXT	Extensions read as text without asking anyone.

Fields inherited from class AbstractTool

input

Fields inherited from class AbstractJob

timeout, upstreamRetries

Fields inherited from interface Job

DEFAULT_UPSTREAM_RETRIES

Fields inherited from interface Tool

OBS_INPUT, OBS_OUTPUT

Constructor Summary

Constructors

Constructor	Description
DeterministicExtractTool(Identifiable parent)

Method Summary

Modifier and Type	Method	Description
Extraction	execute(JobResources resources, JobContext<Extraction> context)	Executes the job with the provided resources and context.
static String	extension(String path)
JobRequirements	getRequirements()	Default requirements for tools.
static String	officeText(Path file)	The text of an OOXML document (docx, xlsx, pptx) with its tags stripped, as readOffice(Path, Extraction) reads it; for any reader of Office files that needs no POI.

Methods inherited from class AbstractTool

getInput, getJobType, postExecute, preExecute, setInput

Methods inherited from class AbstractJob

getId, getName, getParentJobId, getPriority, getTimeout, getUpstreamRetries, getUserId, getWorkflowId, randomIdSuffix, setTimeout, setUpstreamRetries, toString

Methods inherited from class Object

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

Methods inherited from interface GuardedExecution

declareAdmissionGuardrails, declareAuthGuardrails, declareContentGuardrails

Methods inherited from interface Identifiable

getParentJobId, getUserId, getWorkflowId

Methods inherited from interface Job

getDisplayName, getId, getJobId, getName, getPriority, getTimeout, getUpstreamRetries

Methods inherited from interface Tool

guardedInputTarget

Field Details

TEXT

public static final Set<String> TEXT

Extensions read as text without asking anyone.

Constructor Details

DeterministicExtractTool

public DeterministicExtractTool(Identifiable parent)

Method Details

extension

public static String extension(String path)

getRequirements

public JobRequirements getRequirements()

Description copied from class: AbstractTool

Default requirements for tools. Subclasses should override to add database, LLM, or other requirements. JobType is handled by AbstractTool.getJobType(), not here.

Specified by:

getRequirements in interface Job<Extraction>

Overrides:

getRequirements in class AbstractTool<FileRef, Extraction>

Returns:

resource requirements, or null if the job needs no resources

execute

public Extraction execute(JobResources resources,
 JobContext<Extraction> context)
                   throws LLMReadableCheckedException

Description copied from interface: Job

Executes the job with the provided resources and context. A job reports progress and checks for cancellation through the context, and completes all its work before returning. A tool narrows this contract to the LLM-readable exceptions of Tool.execute(JobResources, JobContext).

Parameters:

resources - the allocated resources

context - the job execution context for monitoring

Returns:

the job result, which may be null

Throws:

LLMReadableCheckedException - if execution fails

officeText

public static String officeText(Path file)
                         throws IOException

The text of an OOXML document (docx, xlsx, pptx) with its tags stripped, as readOffice(Path, Extraction) reads it; for any reader of Office files that needs no POI.

Throws:

IOException