agentic-doc-parse-and-extract

🚀 About Laiye ADP

ADP is Laiye's intelligent agent document processing product (Agentic Document Processing, referred to as ADP) , based on the general understanding ability of large models, without relying on rules and annotations, with the general understanding ability of multi-language, MultiModal Machine Learning, and multi-scene; autonomous planning and execution of intelligent agents, able to understand task goals, autonomous planning steps, invoke tools, and complete complex tasks; end-to-end business automation, from document input to business decision-making to human-machine collaboration, forming a complete closed loop.

agentic-doc-parse-and-extract is the official open-source CLI tool of ADP, supporting both manual terminal invocation and automatic invocation via AI Skill. With a single command, it can accomplish: structured document parsing + intelligent extraction of key fields, covering all scenarios including invoices, orders, certificates, bills, and general documents, outputting standard JSON, and seamlessly integrating with automation and AI workflows.

---

💡 Core Features

agentic-doc-parse-and-extract focuses on intelligent processing of the entire document workflow, taking into account both manual terminal calls and automatic calls by AI Agents. Its core functions cover all scenarios of parsing, extraction, and batch processing, requiring no complex configuration, and operations can be completed with a single command:

Installation and Update

Install from Source Code

pip install -e .

Install from PyPI

pip install agentic_doc_parse_and_extract

Version Update

Note: This package is no longer maintained on PyPI. Please update via npm instead:

npm install -g @laiye-adp/agentic-doc-parse-and-extract-cli

Configure

Get an API key at https://adp-global.laiye.com/ (new users get 100 free credits per month).

adp config set --api-key <your-api-key>
adp config set --api-base-url https://adp-global.laiye.com
adp config get

Quick Examples

# List available apps
adp app-id list

# Parse a local document
adp parse local ./invoice.pdf --app-id <app-id>

# Extract key fields
adp extract local ./invoice.pdf --app-id <app-id>

# Parse a directory in async mode
adp parse local ./documents/ --app-id <app-id> --async

# Process a remote URL
adp extract url https://example.com/file.pdf --app-id <app-id>

# Query an async task
adp parse query <task-id>

# Two-phase async (submit + query separately, resumable)
adp extract local ./documents/ --app-id <app-id> --async --no-wait --export tasks.json
adp extract query --watch --file tasks.json

# Auto retry on failure (up to 2 retries)
adp parse local ./documents/ --app-id <app-id> --retry 2

# Check remaining credits
adp credit

Commands

AI agents should call adp schema for the machine-readable, authoritative command spec. The table below is a human-friendly summary.

Flags

Async Workflow

For large files or batch jobs, submit with --async and the CLI returns a task-id. Poll for results with parse query / extract query:

adp parse local ./big.pdf --app-id <app-id> --async
# returns a task-id

adp parse query <task-id>

Two-Phase Async (--no-wait)

By default, --async submits and polls until completion — ideal for AI agents. For resumable workflows, use two-phase mode:

Phase 1: Submit tasks

adp extract local ./documents/ --app-id <app-id> --async --no-wait --export tasks.json

Output is a JSON array with task IDs:

[
  {"path": "invoice.pdf", "task_id": "task_abc123"},
  {"path": "contract.pdf", "task_id": "task_def456"}
]

Phase 2: Query results

adp extract query --watch --file tasks.json
adp extract query --watch --file tasks.json --export ./results/

Even if the CLI crashes mid-way, task IDs in tasks.json are preserved — resume anytime with query --file.

Batch Processing

When processing multiple files/URLs, the CLI writes each result to a separate file:

adp_results_20250417_153020/
├── _summary.json              # Summary (total, success, failed, per-file status)
├── invoice_01.pdf.json        # Successful result
├── contract_02.docx.json
└── report_03.pdf.error.json   # Error details

• --export <dir> — specify output directory
• Without --export — auto-creates adp_results_<timestamp>/
• Single file — outputs to stdout or the --export file path

Exit Codes

Environment Variables

Config Storage

• Config dir: ~/.adp/
• Config file: ~/.adp/config.json
• Encrypted API key: ~/.adp/key.enc (AES-256-GCM)
• App cache: ~/.adp/app_cache.json
• Version check cache: ~/.adp/version_check.json (refreshed every 24h)

📜 License

We adopt a combined model of open-source tools + paid services: the CLI tool is completely free and open-source, making it easy for everyone to quickly integrate; while the core ADP intelligent parsing capability is a Public Cloud commercial service, billed based on actual usage, aiming to provide users with a highly accurate and stable document processing experience.

• CLI Tool: Open source under the MIT License, freely available for use, modification, and distribution
• ADP Service: AI document processing service based on Public Cloud, billed by usage, Billing Rules

Free Quota: New users can receive 100 free credits per month after registration, allowing them to experience full functionality

📞 Support and Contact

• CLI Documentation: ADP CLI User Guide
• API Documentation: OpenAPI User Guide
• User Guide: Public Cloud Operation Manual
• Problem Feedback: GitHub Issues | global_product@laiye.com
• Official Website: Laiye Technology