crewai-xberg

CrewAI tools backed by Xberg. Give an agent document intelligence: extract text, metadata, keywords, entities, and summaries from 107 file formats — PDF, DOCX, XLSX, HTML, images with OCR, and more. Extraction is async at the core; the batch tool routes many files through Xberg's extract_batch in a single native call.

Install

pip install crewai-xberg

Requires Python 3.10+.

Tools

Use

from crewai import Agent
from crewai_xberg import XbergExtractTool, XbergExtractBatchTool, XbergExtractMetadataTool

agent = Agent(
    role="Document Analyst",
    goal="Extract and analyze document content",
    backstory="You process documents of any format.",
    tools=[XbergExtractTool(), XbergExtractBatchTool(), XbergExtractMetadataTool()],
)

Call a tool directly to see its output:

tool = XbergExtractTool()

# Plain extraction
text = tool.run(file_path="report.pdf")

# Force OCR and surface keywords, entities, and a summary
enriched = tool.run(
    file_path="scan.pdf",
    output_format="markdown",
    force_ocr=True,
    extract_keywords=True,
    extract_entities=True,
    summarize=True,
)

# Many files in one batched extraction
combined = XbergExtractBatchTool().run(file_paths=["report.pdf", "notes.docx", "sheet.xlsx"])

Options

Both extraction tools accept the same options. Each toggles an Xberg ExtractionConfig capability:

Detected languages and tables are surfaced automatically when present.

Async

The tools are async at the core. Inside an event loop, await tool.arun(...); the synchronous run bridges to it and must not be called from a running loop.

Errors

A missing file raises from Xberg directly. In batch mode, per-file failures land in ExtractionResult.errors and are reported in a trailing Errors section instead of aborting the batch.

For the full API, see the Xberg documentation.