Document Automation Workspace

Document Automation Workspace is a local-first document automation prototype for turning semi-structured files into reviewed, exportable data.

It combines document upload, schema-based key information extraction, document classification, required-field checks, batch execution, a visual workflow builder, and export jobs in one workspace.

Highlights

• Document library with upload, folder movement, copy, delete, and selection workflows.
• Key information extraction with editable schemas, field regions, AI schema recommendations, field-level review, and JSON/CSV/XLSX export.
• Document classifier and required-field checker modules.
• Workflow Builder with document input, classifier, branch, KIE, required check, merge, and export nodes.
• AI workflow draft generation from up to 10 sample images.
• Inline schema editing inside Workflow Builder KIE nodes before saving the workflow.
• Batch and workflow run monitoring with retry, pause, cancel, result view, and export history.
• Mock VLM mode for local UI and smoke testing without external API calls.

Quick Start

Requirements:

• Python 3.11
• Node.js 20+
• LibreOffice for DOCX/PPTX/XLSX conversion

Run both backend and frontend:

./scripts/run_dev.sh

Then open:

http://127.0.0.1:5173/

The default local configuration can run with VLM_PROVIDER=mock. To call a real provider, copy .env.example to .env and set VLM_PROVIDER, VLM_API_KEY, and VLM_MODEL_NAME.

Useful Commands

Backend tests:

cd backend
../.venv/bin/python -m pytest tests

Frontend build:

npm run build --prefix frontend

Large mock smoke:

./.venv/bin/python scripts/run_large_mock_smoke.py --count 1000

PoC UI smoke:

./.venv/bin/python scripts/run_poc_ui_smoke.py

Public Scope

This public repository focuses on the document automation product surface: upload, extraction, classification, validation, workflow building, execution, review, and export.

Service-only concerns are intentionally not included here.