v0.3-dev

What's New in 0.3 (2025)

• Simpler usage: -i now also analyzes the relevant page text. Using -t together with -i is no longer necessary. Existing -ti calls still work (redundant).
• Faster and cheaper runs: fewer model requests and a smoother live status board.
• Predictable per‑file token limit via [AI].token_limit (default 1,000,000). If the limit is reached, the tool trims lower‑value context first and may skip low‑signal images; INFO logs indicate when this happens.
• No config changes required: current setups continue to work. Tip: adjust [AI].token_limit to trade quality vs. speed/cost.

Includes improvements from 0.2:

• OCR (Tesseract) for scan/low‑text PDFs.
• LiteLLM multi‑provider support (OpenAI tested).
• Parallel job execution with a live status board.
• 24h caching with optional --no-cache and cost reporting.

v0.2-dev

Highlights

• Tesseract OCR integration for robust handling of scans/low text.
• LiteLLM integration enabling multiple providers and local models (OpenAI tested).
• Parallel job execution with a live status board (OCR, image, text).
• Simplified image prioritization: early pages first, prefer larger images/scans, group tiny icons; per‑PDF cap.

Image Analysis

• Page‑local context: OCR/text from the same page is attached to each image.
• Groups icon clusters into a page render instead of many tiny images.
• Ordering fix: images → inject alt‑texts into page text → run text analysis.

Caching & Costs

• 24h on‑disk cache for OCR, chat, and vision; disable with --no-cache.
• Cost reporting shows both spent and saved (via cache).
• Tag analysis made cache‑friendly by sorting/deduplicating tags before requests.

Export & Metadata

• Originals remain untouched; metadata is written only to exported copies.
• Writes standard fields: Title, Subject (Summary), Author, Keywords, CreationDate.
• Configurable filename scheme via [EXPORT].filename_format (strftime + {TITLE}/{CREATOR}) with automatic slugification.

Defaults & Configuration

• Models: short text → openai/gpt-5-mini; long text & images → openai/gpt-5-nano.
• New config knobs under [AI] (e.g., max_images_per_pdf, first_pages_priority, scan_coverage_threshold), [CACHE], and [EXPORT].
• Cache directory defaults to ~/.autoPDFtagger/cache.

CLI/UX

• Quieter third‑party logs to keep the status board readable.
• Image selection summary logs how many images per page were used.

Tests & Quality

• Expanded unit/integration tests for OCR, LLM client, image selection, CLI, export, and caching.
• Hardened export path with clear error logs and a final “Export summary”.

Upgrade Notes

• Install Tesseract (system package) and copy the sample config:
  • cp autoPDFtagger_example_config.conf ~/.autoPDFtagger.conf
  • Set provider API keys (e.g., OPENAI_API_KEY).
• Optionally customize [EXPORT].filename_format and [CACHE] settings.

Known Issues

• Costs may show 0.00 for models without known pricing unless [PRICING] config is provided.
• Ensure the cache directory is writable; on failure, caching degrades gracefully.

Bug fixes and CSV-Import/Export function

What's Changed

• Dev by @Uli-Z in #3

Full Changelog: v0.1.1-dev...v0.1.2-dev

v0.1.1-dev

Initial release.