AGENTFlow is a Python library for building agent-oriented applications with a small, composable surface area.

It currently focuses on three practical integration layers:

• agentflow.connectors: lightweight connectors such as arXiv
• agentflow.llms: stable wrappers around model gateways such as LiteLLM
• agentflow.parsers: document parsing wrappers such as MinerU

The project is designed to be imported from other Python projects, not only run as a local app scaffold.

Install the base package when you only need dependency-light modules such as the arXiv connector:

pip install agentflow

Install optional integrations only when you need them:

pip install "agentflow[llms]"
pip install "agentflow[mineru]"
pip install "agentflow[mineru-pipeline]"

For local development with uv:

uv sync
uv sync --extra llms
uv sync --extra mineru-pipeline

LiteLLM and MinerU are now optional dependencies. That means:

• import agentflow works even if litellm or mineru is not installed.
• users can install only the integrations they actually need.
• the package is better suited for publishing on PyPI as a reusable library.

If you access an optional feature without its dependency installed, AGENTFlow raises a clear installation hint.

from agentflow import ArxivConnector

connector = ArxivConnector()
paper = connector.resolve("1706.03762")
print(paper.pdf_url)

from agentflow import LiteLLMClient, LiteLLMConfig

client = LiteLLMClient(
    LiteLLMConfig(
        model="gpt-4o-mini",
        api_key="your-api-key",
    )
)

response = client.prompt("Say hello in one short sentence.")
print(response.content)

from agentflow import MinerUConfig, MinerUParser

parser = MinerUParser(
    MinerUConfig(
        backend="pipeline",
        parse_method="auto",
        lang_list=("ch",),
    )
)

result = parser.parse_file("example.pdf")
print(result.parse_dir)
print(result.markdown_file)

from pathlib import Path

from agentflow import MinerUParser

parser = MinerUParser()
pdf_bytes = Path("example.pdf").read_bytes()
result = parser.parse_bytes(pdf_bytes, file_name="example.pdf")
print(result.middle_json_file)

Minimal test scripts live next to the modules they exercise:

• src/agentflow/connectors/test_arxiv_connector_minimal.py
• src/agentflow/llms/test_litellm_minimal.py
• src/agentflow/parsers/test_mineru_minimal.py

Examples:

python src/agentflow/connectors/test_arxiv_connector_minimal.py --search transformer
python src/agentflow/parsers/test_mineru_minimal.py your.pdf

• MinerUParser now uses direct local invocation instead of spinning up a temporary mineru-api service.
• MinerUParser.parse_bytes() and MinerUParser.aparse_bytes() are intended to make downstream library integration easier.
• the top-level package uses lazy exports so optional integrations do not break unrelated imports.

Before uploading to PyPI, you should still confirm:

• the package name agentflow is available on PyPI
• your final project license choice
• the repository/homepage URLs you want to publish in package metadata

Build commands:

python -m build
python -m twine check dist/*
python -m twine upload dist/*