v0.0.20

• Added ChatGPT (3.5 tubo) model support, which cuts costs by 10x
• Made it so citations are guessed now (can still be set manually)
• Various prompt tweaks to account for new model

v0.0.19

• Fixed bug on including contexts that said Not Applicable (thanks @ljch2018 )
• Shortened prompts and added more guards on inapplicable - These new prompts will change answers
• Added default caching in a SQL file in home directory (thanks @xloem)

v0.0.18

Fixed truncation on completions longer than 100 words

v0.0.17

Made recent years more flexible

v0.0.16

• Fixed bug in dense PDFs (one page cannot fit into chunk)

v0.0.15

• Reverted release with removal of dependency

v0.0.14

• Made PyCyptodome optional because of its size

v0.0.13

• Fixed overlap bug affecting short documents
• More options in query generation
• Made generator version to get intermediate feedback

v0.0.12

• Improved pickling
• Made it possible to update llms
• New function for proposing searches
• Improved prompts

v0.0.10

Added token usage and estimated price to answers