November 24: Support for direct LLM prompt, multi-turn image analysis, bug fixes

New Features

• 💡 Graphlit now supports multi-turn analysis of images with the reviseImage and reviseEncodedImage mutations. You can provide an LLM prompt and either a URI or Base-64 encoded image and MIME type, along with an optional LLM specification. This can be used for analyzing any image and having a multi-turn conversation with the LLM to revise the output from the LLM. (Colab Notebook Example)

• 💡 Graphlit now supports directly prompting an LLM with the prompt mutation, bypassing any RAG content retrieval, while providing an optional list of previous conversation messages. This also accepts an optional LLM specification. (Colab Notebook Example)

• We have added support for the new Mistral Pixtral Large model, with PIXTRAL_LARGE model enum, which can be used with LLM completion or entity extraction LLM specifications.

• We have added support for the OpenAI 2024-11-20 version of GPT-4o, with GPT4O_128K_20241120 model enum.

• ⚡ We have added Microsoft Entra ID (fka Azure Active Directory) clientId and clientSecret properties to the SharePointFeedPropertiesInput type, which are now required when creating a SharePoint feed using user authentication with refreshToken property. (Colab Notebook Example)

Bugs Fixed

• GPLA-3438: Not filtering on desktop presentation when scraping web pages

• GPLA-3340: Failed to parse invalid JSON from extracted PDF page

• GPLA-3427: Not formatting extracted tables properly from Sonnet 3.5