Offline Newsletter Builder (Archaeology/History)

Local-first toolchain to turn your article links (or pasted texts) into a Friday Ghost draft with consistent summaries, plus a SQLite database for memory and related-article suggestions.

Highlights

• Paste URLs into inbox.txt → run importer → get .md stubs and a growing SQLite DB.
• One command to build the weekly newsletter (HTML) and optionally create a Ghost draft.
• Consistent tone via templates/prompt.txt (+ your templates/style_bank.md samples).
• "Memory" via data/newsletter.db (sources, summaries, embeddings, FTS), used to auto-suggest related items.

Quick start

python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

# Configure secrets
cp .env.example .env
# Edit .env to set your keys (or point to local OpenWebUI/Ollama)

# Create DB schema
python scripts/db_cli.py init

# Ingest URLs (from inbox.txt)
python scripts/ingest_list.py --list inbox.txt --fetch --stubs --date 2025-09-19 --sleep 0.5

# Build newsletter from stubs (HTML + optional Ghost draft)
python build.py entries/2025-09-19 --out dist/2025-09-19.html --publish

# (Optional) Build directly from DB (Top-N sources)
python scripts/build_from_db.py --limit 10 --out dist/2025-09-19.html --publish

Sample Run (Pretend Output)

$ make init
Initialized schema in data/newsletter.db

$ make stubs DATE=2025-09-19
Stub: entries/2025-09-19/bbc-welsh-rainforests.md
Stub: entries/2025-09-19/nature-bone-discovery.md
Ingested: 2 URLs into data/newsletter.db

$ make build DATE=2025-09-19
[build] Using LLM provider=openwebui model=qwen2.5-7b-instruct
[build] Generating summary for: Velšské deštné pralesy ...
[build] -> Summary written back to DB (id=5, embedding stored)
[build] Generating summary for: Kosterní nález u Nisy ...
[build] -> Summary written back to DB (id=6, embedding stored)
Saved: dist/2025-09-19.html

$ make draft DATE=2025-09-19
Draft: https://your-ghost-site.ghost.io/p/objevy-tydne-2025-09-19-draft

You can open the generated HTML under dist/2025-09-19.html in your browser to review before publishing.