0.8.0

• [invalidating] Add a filter step to the workflow to filter condensed articles by section before they are used to write a section. This was added to give more of a chance for all articles to be used when there are more articles than can fit in the LLM context. As for when there aren't too many articles, the output should nevertheless be more focused. This will invalidate the preexisting disk cache after this step.
• Update the citation pattern from [1,2,3] to [1][2][3] for easier searching. This affects these output formats: txt, md, html, pdf
• Update the user agent string as its prior value was no longer working.
• Update the workflow to remove any empty sections, if any, at each applicable step.

0.7.2

• Mitigate a diskcache hang by limiting the rate_articles and condense_articles modules to use a max of 12 threads instead of 16.

0.7.1

• [invalidating] Update the condense_article and combine_articles prompts to perform a critical assessment. This will negate the reuse of prior disk cache.
• Update all sample links in readme to use GitHub Pages links for more reliable HTML rendering.

0.7.0

• [invalidating] If the user query is rejected by the LLM for any reason, the rejection reason is now printed as an error. This negates prior disk cache for the LLM as the prompt has changed.

0.6.1

• Update HTML output to use Tippy.js to show tooltip. This fixes issues with the tooltip's placement.

0.6.0

• [invalidating] Update list_search_terms prompt to reject invalid input such as a file path, etc. This change to the prompt however negates the prior disk cache.
• Update combine_articles module to add a validation check to ensure that there is at least one citation in each section for which one or more articles are available.

0.5.1

• Update the lower limit of max_sections from 10 to 5.

0.5.0

• [invalidating] The number of disk cache shards were increased from 8 to 16. This new value should better support the 16 concurrent OpenAI workers. This will negate the prior disk cache.
• Add support for physorg source for science articles. It is documented in the readme. Corresponding sample files are also linked in the readme.
• Various minor improvements.

0.4.2

• Make minor improvements to PDF output.
• Fix condense_articles function to limit max_workers to 8 in order to work around diskcache hanging if this number is higher.

0.4.1

• Support PDF output using reportlab.
• Update openai to 1.45.0 (from 1.44.1) and also other dependencies.