Last updated
Graphlit now supports mapping a website with the mapWebmutation. You can provide a URL to a website, and the query will return a list of URLs based on the sitemap.xml (or sitemap-index.xml) file, at or underneath the provided URL.
Graphlit now supports the generation of web page screenshots with the screenshotPagemutation. By providing the URL of a web page, and optionally, the maximum desired height of the screenshot, we will screenshot the webpage and ingest it automatically as content. You can provide an optional workflow, which will be applied to the ingested image content, for operations like generating image descriptions with a vision LLM.
Graphlit now supports the direct summarization of text with the summarizeTextmutation. By providing the desired summarization strategy, we will summarize the text (i.e. bullet points, social media posts) and return the summarization.
Graphlit now supports the direct extraction of text with the extractTextmutation. By providing the LLM tool definitions and an optional LLM specification, we will prompt the desired LLM (or OpenAI GPT-4o, by default) to invoke the provided tools, and return the JSON responses from the LLM tool calling.
Graphlit now supports the latest Groq Llama 3.3 model, with the model enum LLAMA_3_3_70B.
We have updated Cohere reranking to use the latest Cohere rerank-v3.5model by default.
We have added a new flattenCitationsfield to the ConversationStrategyInputtype. By assigning this field to True, when calling promptConversation,we will combine multiple citations from the same content into a single citation.
For Microsoft email, Microsoft Teams and OneDrive feeds, we have added the clientIdand clientSecretfields as required feed properties. These properties must be assigned, in addition to the refreshTokenfield for proper authentication to the Microsoft Graph API used by these feeds. ()
GPLA-3492: Not finding sitemap at parent web path
GPLA-3500: Failed to handle mismatch of Deepgram model/language
