v3.3.2

• Changed prompt collection type to use default initializers to allow json encoding

v3.3.1

• Made query vs question vs answer consistent in aquery

v3.3.0

What's Changed

• More improvements for reducing memory usage by @whitead in #160

Full Changelog: v3.2.1...v3.3.0

v3.2.1

• Improved capturing of relevance score to account for wider response formats

v3.2.0

• Made it possible to have doc_index entirely empty (provided by external vector DB)

v3.1.1

What's Changed

• Fix the arguments for a function call to add_embeddings by @thenatefisher in #147
• Add unit tests and version bump for add_text fix by @whitead in #151

New Contributors

• @thenatefisher made their first contribution in #147

Full Changelog: v3.1.0...v3.1.1

v3.1.0

add, add_file, and add_url now return None if the doc to add was a duplicate. add_texts now return False if there was a dulicate.

v3.0.0.dev3

• Added memory model

v3.0.0

Breaking changes in v3

See README

Release Notes:

• Added memory model
• Changed variable naming to be much more clear
• More types to avoid ambiguous dictionary/tuples
• Switched to using pydantic instead of dataclasses
• Removed non-core focus elements (caching, token counting, agents)
• Removed need to have texts/doc names in Docs, removing RAM problems
• Abstracted FAISS away, so any vector store can be used
• Removed ambiguous chunk/overlap defaults
• Added typewriter example to README
• Added caching example to README
• Made prompting more general
• Can pass by FileObj|Path|URL in add - (add, add_file, add_url)
• Reduce code duplication in unit tests
• Added more flexible typing, exported more types, added back reader default args

Contributors: @whitead and @Mason-McGough

v3.0.0.dev2

• Added more flexible typing, exported more types, added back reader default args