Backend migration:
- Replace pyodbc/SQL Server with psycopg2/PostgreSQL throughout
- Rewrite Database class with portable SQL: SERIAL, ON CONFLICT, NOW()
- Lowercase table names (rip_help_files, rip_help_sections) - Postgres convention
- libpq connection string format in HELP_DB_CONN
Webapp (webapp/):
- FastAPI app: GET /, GET /images/<f>, GET /home-image, GET /api/sections,
POST /api/keywords/<code>, GET /healthz
- Jinja2 template extracted from generate_html.py with HTTP image URLs
- Direct keyword save to DB (no JSON download detour)
- Same prefix scoping as CLI tools (?prefix=RIP)
Deployment:
- Dockerfile (python:3.12-slim + uvicorn)
- docker-compose.yml for local dev
- requirements-webapp.txt (minimal, no Windows-only deps)
- .dockerignore excludes pipeline scripts and BAT files
- README updated with webapp section and Coolify deploy guide
Also: switch AI model to claude-haiku-4-5 (~3x cheaper, same quality for this task)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- help_processor.py: parses .docx/.html/.pdf/.doc/.txt, extracts images,
classifies sections via Claude API, writes to SQL Server
- generate_html.py: builds interactive HTML viewer (Home/Editor/Search/Generator)
- save_keywords.py: applies keyword edits back to DB
- Prefix-scoped DB schema (RIP_help_files, RIP_help_sections) so multiple
projects share the same database without collision
- BAT launchers per project (RIP_load.bat, INEX_TM_load.bat, ...) load
credentials from gitignored .env via _load_env.bat
- Rich HTML preservation for .html sources (html_text column)
- Image extraction for all formats with MS Word / LibreOffice fallback for .doc
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
