March 13: Support for classification workflow, notifications, Cohere Command A model, bug fixes

New Features

• 💡 Graphlit now supports a classification stage in the workflow object. As content is being ingested, and after the extraction stage in the workflow, you can optionally enable classification rules to apply labels to the content based on regex patterns or LLM prompts. For example, an LLM prompt classification rule can use Gemini 2.0 Flash to analyze the Markdown text of an email, and assign the 'Invoice' label to the content automatically, if the prompt determines the email contains an invoice to be paid.

• Graphlit now supports the Cohere Command A model, with the model enum COMMAND_A.

• We have added notification mutations to send textual notifications to Slack, email or webhook. The new sendNotificationmutation is used in MCP tools to directly send the Markdown output from a tool to Slack, email or webhook.

Bugs Fixed

• GPLA-3961: Failed to extract hyperlinks from Word document

• GPLA-3946: Google Drive feed is ingesting a duplicate rather than in-place

• GPLA-3905: Slow to load JSON mezzanine for CSV

• GPLA-3966: If mapWeb doesn't find anything, just return original URL

• GPLA-3959: Failed to process hyperlink spreadsheet cell