CHANGELOG.md

Changelog

0.2.0 — 2025-10-24

Highlights

• Local and multi‑provider LLMs: In addition to OpenAI, you can now use local models via Ollama and other providers like Gemini through LiteLLM.
• Simple AI config: Choose explicit models per task in [AI] — no code changes needed.
• Clearer cost logging: Costs are estimated from API usage and reported per analysis.

What’s new

• New config keys under [AI]:
  • text_model_short, text_model_long, and text_threshold_words for short/long text routing
  • image_model for vision/image analysis
  • tag_model for tag consolidation
  • image_temperature (default 0.8) for vision responses
• API key fallback: If OPENAI_API_KEY is not set, the app can read [OPENAI-API].API-Key from your config as a fallback.

Smoother architecture

• The old OpenAI‑specific agent classes were removed. A thin, provider‑agnostic LiteLLM wrapper is used instead. Your setup happens in the config, not the code.

Migration guide (3 steps)

1. Update dependencies (ideally in a fresh virtualenv):
   • pip install -e .
2. Update your config (~/.autoPDFtagger.conf):
   • Under [AI], set the models you want. Examples:
     • text_model_short = openai/gpt-4o
     • text_model_long = openai/gpt-4o-mini
     • text_threshold_words = 100
     • image_model = openai/gpt-4o (or gemini/gemini-1.5-pro or ollama/llava)
     • tag_model = openai/gpt-4o-mini
     • image_temperature = 0.8
   • API key: Prefer environment variable OPENAI_API_KEY. As a fallback, you can put it under [OPENAI-API] → API-Key = sk-....
3. Try it:
   • autoPDFtagger ./pdfs -f -t -i -c --json ./out.json

Notes

• If image analysis is skipped, ensure image_model is set and supports vision.
• Costs are logged at the end of each analysis phase.