Output Layer

Output Layer

OpenBench output layer generates artifacts from processed data.

Key Files

• src/openbench/output/generators.py - All generator implementations
• src/openbench/output/layer.py - OutputFactory
• src/openbench/core/abstractions.py - OutputGenerator base class, GeneratedOutput

Available Generators

PDFGenerator

from openbench.output.generators import PDFGenerator

generator = PDFGenerator(
    template="report",        # "default", "report", "corporate"
    page_size="letter",       # "letter", "a4", "legal"
    font_name="Helvetica",
    font_size=11,
    title_font_size=18,
    heading_font_size=14,
)

result = generator.generate(
    content="Report content here...",
    output_path="report.pdf",
    title="Q1 Report",
)
# Returns GeneratedOutput(file_path, format, size_bytes, metadata)

Content types supported:

• str - Rendered as paragraphs
• dict - Keys as section headings, values as content
• list - Rendered as bullet points

MarkdownGenerator

from openbench.output.generators import MarkdownGenerator

generator = MarkdownGenerator()
result = generator.generate(content=data, output_path="output.md")

Using in Workflows

from openbench.core import OutputLayer
from openbench.output.generators import PDFGenerator, MarkdownGenerator

# Single output
workflow = data | intelligence | OutputLayer(generators=PDFGenerator())

# Multiple outputs in parallel
workflow = data | intelligence | OutputLayer(
    generators=PDFGenerator() & MarkdownGenerator()
)

OutputGenerator Interface

from openbench.core.abstractions import OutputGenerator, GeneratedOutput

class MyGenerator(OutputGenerator):
    @property
    def output_format(self) -> str:
        return "custom"  # Required

    def generate(self, content, template=None, **options) -> GeneratedOutput:
        file_path = self._create_output(content)
        return GeneratedOutput(
            file_path=file_path,
            format=self.output_format,
            size_bytes=os.path.getsize(file_path),
            metadata={"template": template}
        )

    def validate(self, content) -> bool:
        return content is not None  # Required

Anti-Patterns

DO NOT:

• Import heavy dependencies at module level (reportlab, python-pptx) - use lazy imports
• Assume content is always a string - check type and handle dict/list/str
• Forget validate() - it's called before generate() in the pipeline
• Return raw file paths - always return GeneratedOutput dataclass
• Skip output_format property - it's required by the abstract class

Cross-References

• Composing Workflows: Generators are Chainable, usable with | & in OutputLayer → see composing-workflows skill
• Creating Abstractions: OutputGenerator base class details → see creating-abstractions skill
• Intelligence Layer: Agent output feeds into generators → see intelligence-layer skill

For examples, see examples/workflows/reports/sustainability_report.py

Converted and distributed by TomeVault — claim your Tome and manage your conversions.