Implementation:Neuml Txtai TemplateTask

Knowledge Sources	Neuml_Txtai
Domains	Workflow, Templating, LLM Prompting
Last Updated	2026-02-10 01:00 GMT

Overview

Concrete tool for generating text from templates and task inputs within workflows provided by txtai.

Description

The TemplateTask class extends the base Task class to generate text by applying Python format string templates to input data. It supports three input modes:

• dict input: Template placeholders are filled from dictionary keys (e.g., {name} maps to element["name"])
• tuple input: Template placeholders use positional arguments ({arg0}, {arg1}, etc.)
• string input: The element is substituted into a {text} placeholder

The class also supports processing rules, which allow short-circuiting the template when a dict element matches a rule condition. In strict mode (default), all template placeholders must be consumed by the input; non-strict mode uses Python's standard Formatter which ignores missing keys. The module also includes RagTask, a subclass that prepares input for RAG pipelines by separating query and question fields.

Usage

Use TemplateTask in txtai workflows when you need to construct formatted text from structured data, such as generating LLM prompts, preparing input for downstream pipeline stages, or transforming data between workflow steps. It is particularly valuable for RAG workflows where questions need to be formatted with context before being sent to a language model.

Code Reference

Source Location

• Repository: Neuml_Txtai
• File: src/python/txtai/workflow/task/template.py

Signature

class TemplateTask(Task):
    def register(self, template=None, rules=None, strict=True)
    def prepare(self, element)
    def defaulttemplate(self)
    def defaultrules(self)
    def match(self, element)

class RagTask(TemplateTask):
    def prepare(self, element)

Import

from txtai.workflow.task.template import TemplateTask, RagTask

I/O Contract

Inputs

Name	Type	Required	Description
elements	list	Yes	List of data elements (str, dict, or tuple) to apply the template to
executor	object	No	Optional executor instance for concurrent task actions
template	str	No (register)	Python format string template with placeholders; defaults to None (passthrough)
rules	dict	No (register)	Dictionary of key-value rules for conditional short-circuit matching; defaults to empty dict
strict	bool	No (register)	If True (default), all template placeholders must be consumed; if False, missing keys are ignored

Outputs

Name	Type	Description
results	list	List of formatted text strings generated by applying the template to each input element
results (RagTask)	list[dict]	List of dicts with "query" and "question" keys, where "question" has the template applied

Usage Examples

from txtai.workflow.task.template import TemplateTask, RagTask

# String input with {text} placeholder
template = TemplateTask(
    template="Summarize the following: {text}",
    action=lambda x: x
)
results = template(["This is a long document about AI."])
# Returns ["Summarize the following: This is a long document about AI."]

# Dict input with named placeholders
template = TemplateTask(
    template="Question: {question}\nContext: {context}",
    action=lambda x: x
)
results = template([{"question": "What is AI?", "context": "AI stands for..."}])
# Returns ["Question: What is AI?\nContext: AI stands for..."]

# Tuple input with positional arguments
template = TemplateTask(
    template="{arg0} is related to {arg1}",
    action=lambda x: x
)
results = template([("Python", "programming")])

# RAG task for preparing retrieval-augmented generation input
rag = RagTask(
    template="Answer this question: {text}",
    action=lambda x: x
)
results = rag(["What is machine learning?"])
# Returns [{"query": "What is machine learning?",
#           "question": "Answer this question: What is machine learning?"}]

# Using rules for conditional processing
template = TemplateTask(
    template="Process: {text}",
    rules={"type": "skip"},
    action=lambda x: x
)
results = template([{"type": "skip", "text": "should be skipped"}])
# Returns ["skip"] (short-circuited by rule match)

Related Pages

• Environment:Neuml_Txtai_Python_Core_Dependencies