Added

• Add DALL·E 3 to specs and README - thanks to @Gary-H9
• Add Whisper transcription language selection parameter to README - thanks to @nfedyashev
• Add bundle exec rake lint and bundle exec rake test to make development easier - thanks to @ignacio-chiazzo
• Add link to https://github.com/sponsors/alexrudall when users run bundle fund

Fixed

• Update README and spec to use tool calls instead of functions - thanks to @mpallenjr
• Remove nonexistent Thread#list method - thanks again! to @ignacio-chiazzo
• Update finetunes docs in README to use chat instead of completions endpoint - thanks to @blefev

Fixed

• Allow any kind of file (eg. PDFs) to be uploaded to the API, not just JSONL files. Thank you @stefan-kp for the PR!

Added

• Add ability to pass Faraday middleware to the client in a block, eg. to enable verbose logging - shout out to @obie for pushing for this.
• Add better error logging to the client by default.
• Bump Event Source to v1, thank you @atesgoral @ Shopify!

Added

• Add text-to-speech! Thank you @codergeek121

Added

• Add support for Assistants, Threads, Messages and Runs. Thank you @Haegin for the excellent work on this PR, and many reviewers for their contributions!

Fix

• Gracefully handle the case where an HTTP error response may not have valid JSON in its body. Thank you @atesgoral!

Added

• [BREAKING] HTTP errors will now be raised by ruby-openai as Faraday:Errors, including when streaming! Implemented by @atesgoral
• [BREAKING] Switch from legacy Finetunes to the new Fine-tune-jobs endpoints. Implemented by @lancecarlson
• [BREAKING] Remove deprecated Completions endpoints - use Chat instead.

Fix

• [BREAKING] Fix issue where :stream parameters where replaced by a boolean in the client application. Thanks to @martinjaimem, @vickymadrid03 and @nicastelo for spotting and fixing this issue.

Fix

• Added more spec-compliant SSE parsing: see here https://html.spec.whatwg.org/multipage/server-sent-events.html#event-stream-interpretation
• Fixes issue where OpenAI or an intermediary returns only partial JSON per chunk of streamed data
• Huge thanks to @atesgoral for this important fix!

Added

• Added rough_token_count to estimate tokens in a string according to OpenAI's "rules of thumb". Thank you to @jamiemccarthy for the idea and implementation!

Added

• Support multi-tenant use of the gem! Each client now holds its own config, so you can create unlimited clients in the same project, for example to Azure and OpenAI, or for different headers, access keys, etc.
• [BREAKING-ish] This change should only break your usage of ruby-openai if you are directly calling class methods like OpenAI::Client.get for some reason, as they are now instance methods. Normal usage of the gem should be unaffected, just you can make new clients and they'll keep their own config if you want, overriding the global config.
• Huge thanks to @petergoldstein for his original work on this, @cthulhu for testing and many others for reviews and suggestions.

Changed

• [BREAKING] Move audio related method to Audio model from Client model. You will need to update your code to handle this change, changing client.translate to client.audio.translate and client.transcribe to client.audio.transcribe.