AgenticParser

AgenticParser parses Markdown, PDF, and DOCX files into a unified intermediate representation, applies auditable cleanup rules, and exports clean Markdown with JSON reports.

Features

• Unified IR for documents, pages, blocks, and spans.
• Rule-based cleanup with traceable policy decisions.
• Markdown, PDF, and DOCX input support.
• Page inspection metadata for PDF/DOCX workflows.
• Quality scoring and issue reports.
• Single-file and batch parsing commands.

Installation

Install the package in editable mode:

pip install -e .

For PDF parsing, install the optional PDF dependency:

pip install -e ".[pdf]"

For local source execution without installation:

$env:PYTHONPATH="src"

Basic Usage

Parse one document:

python -m agentic_parser parse `
  path\to\input.md `
  --memory memory\rules `
  --out outputs\latest

Parse a PDF or DOCX file the same way:

python -m agentic_parser parse `
  path\to\input.pdf `
  --memory memory\rules `
  --out outputs\latest

Batch parse a directory:

python -m agentic_parser parse-batch `
  path\to\documents `
  --memory memory\rules `
  --out outputs\batch

Outputs

Each parse run writes:

• parsed.md: cleaned Markdown.
• parsed.ir.json: full intermediate representation.
• report.json: applied rules, issues, backend report, and quality score.

Batch parsing also writes:

• batch_summary.json: status and score summary for each file.

Rules

Rules are Markdown files in memory/rules/ with TOML frontmatter.

Example:

id = "drop_strike_text_contract"
priority = 100
action = "drop_span"
reason = "Drop obsolete revision text."

[scope]
doc_family = "contract"

[trigger]
style = "strike"

Supported actions:

• drop_span
• drop_block

Common trigger fields:

• block_type
• style
• text_contains

Common scope fields:

• doc_family
• template_cluster

CLI Options

Useful options:

• --backend: choose plain-markdown, pdf, or docx; defaults to auto.
• --doc-family: pass a document family used by scoped rules.
• --revision-agent: use none or heuristic.
• --memory: path to rule files.
• --out: output directory.

Python API

from pathlib import Path

from agentic_parser.pipeline import run_parse_pipeline

result = run_parse_pipeline(
    input_path=Path("path/to/input.md"),
    memory_dir=Path("memory/rules"),
    doc_family="contract",
)

print(result.markdown)
print(result.report())

Tests

$env:PYTHONPATH="src"
python -m unittest discover -s tests