memect-ppx

Parse PDF and images into structured Markdown / JSON using PPX. Supports OCR, table extraction, formulas, multi-column layouts, and LLM backends.

Security Scan

The skill's requests and instructions are consistent with a local PDF/image parsing CLI wrapper (ppx); no unexplained credentials, installs, or hidden behaviors are present, but be careful about configuring remote LLM backends which could transmit documents off your machine.

Purpose & Capability

Name/description, required binary ('ppx'), and the SKILL.md usage (ppx CLI, Python API) align: this is a CLI/Python wrapper for memect-ppx that parses PDFs/images into Markdown/JSON. No unrelated binaries or credentials are requested.

Instruction Scope

Instructions stay within the parser's domain (parsing files, saving outputs, optional debug/dev modes). However the skill documents support for configurable LLM backends (base_url and api_key fields) — if a user points those to remote services, their documents may be sent externally. The SKILL.md itself does not instruct reading unrelated files or exfiltrating data, but backend configuration introduces a privacy/egress decision the user must control.

Install Mechanism

No automated install spec is present (instruction-only). SKILL.md recommends pip install memect-ppx and dependencies which is a normal, traceable install method; there are no opaque download URLs or archive extraction steps in the skill bundle.

Credentials

The skill declares no required environment variables or credentials (primaryEnv null). It does show optional api_key fields in backend JSON examples — these are expected for remote LLM backends but are not required by the skill itself. Users should be aware that supplying backend credentials enables networked inference and could expose documents to third parties.

Persistence & Privilege

always is false and the skill is user-invocable; it does not request persistent system-wide privileges or modify other skills. Running the parser will write outputs and optional intermediate files to user-specified directories (expected behavior).

Assessment

This skill appears to be what it claims: a local ppx-based PDF/image parser. Before installing or using it, verify the memect-ppx pip package source (e.g., PyPI package page or upstream GitHub), run it in an isolated environment if you handle sensitive documents, and be careful when configuring LLM backends — pointing base_url to a remote service or supplying an api_key will cause your documents (or extracted content) to be sent to that service. If you need strict privacy, use local backends only (localhost) or avoid configuring external backends.

PPX 解析 Skill

PPX（memect-ppx）是一款本地运行的高精度文档解析引擎，将 PDF / 图片转换为 结构化 Markdown 和 JSON，无需 GPU，开箱即用。

安装

pip install memect-ppx
pip install onnxruntime
pip install opencv-contrib-python

Python ≥ 3.12。在 Linux 无头环境下将 opencv-contrib-python 替换为 opencv-python-headless。

基本用法

# 解析单个 PDF，输出到 output/ 目录
ppx parse report.pdf -o output/

# 解析图片
ppx parse scan.png -o output/

# 批量解析目录
ppx parse docs/ -o output/

# 只解析指定页面
ppx parse report.pdf --pages "1-5,10,15-20" -o output/

输出结构（写入 <file>.out/）：

路径	说明
doc.md	完整文档 Markdown，含图形引用
doc.json	JSON 树：文档 → 页面 → 对象，每个对象含坐标
pages/	逐页 Markdown / JSON，适合页面级处理
images/	提取的图像区域（检测到图形时存在）

OCR 控制

ppx parse report.pdf --ocr auto    # 默认：自动判断
ppx parse report.pdf --ocr yes     # 强制 OCR（扫描件）
ppx parse report.pdf --ocr no      # 跳过 OCR（原生 PDF）

表格解析模式

ppx parse file.pdf --table auto    # 默认：自动选择
ppx parse file.pdf --table wbk     # 白底表格（推荐普通文档）
ppx parse file.pdf --table ybk     # 彩色/深色背景表格
ppx parse file.pdf --table llm     # LLM 辅助（最高精度，需配置后端）
ppx parse file.pdf --table no      # 跳过表格解析

LLM 后端

后端	场景	显存需求
default	隐私、离网、CI	CPU，无需 GPU
deepseek	复杂版面最高精度	~20 GB
paddle	精度较好、显存小	~10 GB
glm	快速推理	~10 GB

# 使用 DeepSeek-OCR-2 后端（需本地 vLLM 服务）
ppx parse report.pdf --backend deepseek \
  --deepseek '{"base_url":"http://127.0.0.1:4000/v1","model":"deepseek-ocr-2","api_key":""}'

# 使用 PaddleOCR-VL 后端
ppx parse report.pdf --backend paddle \
  --paddle '{"base_url":"http://127.0.0.1:4001/v1","model":"paddleocr-vl","api_key":""}'

# 使用 GLM-OCR 后端
ppx parse report.pdf --backend glm \
  --glm '{"base_url":"http://127.0.0.1:4002/v1","model":"glmocr","api_key":""}'

持久化配置（避免每次重复填写）：

ppx parse report.pdf --set backend="deepseek" \
  --set deepseek.base_url="http://127.0.0.1:4000/v1"

其他选项

--mode page|tree|ppt     # 解析模式（默认 page）
--workers N              # 目录并行解析线程数
--docx / --no-docx       # 输出 DOCX
--pptx / --no-pptx       # 输出 PPTX
--json                   # 仅输出 JSON
--cpu                    # 强制 CPU（即使有 GPU）
--debug / -x             # 输出调试信息和中间图
--dev                    # 开发模式，保存中间结果

Python API

from memect.pdf.parser import Parser
from memect.pdf.base import KDocument

# 单次解析
with Parser() as parser:
    doc = KDocument("/path/your.pdf")
    parser.parse(doc)

# 批量多进程
from memect.pdf.base import KDocumentFactory
docs = [KDocumentFactory("/path/your.pdf", params=None)]
Parser.batch(docs, max_workers=4)

Parser 对象全局共享，多次调用时只需实例化一次。

常见问题

• Linux ImportError: libGL.so.1 → pip install opencv-python-headless 或 sudo apt-get install -y libgl1
• Mac 无法 GPU 加速 → Apple Silicon / Intel Mac 不支持 CUDA，CPU 模式正常运行
• onnxruntime 与 onnxruntime-gpu 冲突 → 只安装其中一个

许可证

非商业免费（PolyForm Noncommercial 1.0.0）。商用联系 [e-mail]。 在线体验：https://pdf2x.cn/