April 13: Support for memory, email thread collections, Groq Llama 4 models, bug fixes

New Features

  • 💡 Graphlit now supports MEMORY as a new content type. Via the ingestMemory mutation, you can create a textual memory for your user or agent. (Not supported at project scope.) Memories will automatically have entity extraction applied to create an 'interest graph' from the accumulated memories. Memories will be included with retrieved sources for RAG conversations, via vector similarity. (Coming soon: memories will support time-to-live (TTL), so they will age out automatically.)

  • We have added support for auto-creating collections for email threads. You can opt into this behavior with enableEmailCollections in IngestionWorkflowStage configuration.

  • We have added support for Google Drive service account authentication. You can now provide the JSON credentials from Google Cloud Console to authenticate to a Google Drive feed.

  • We have added support for the latest Llama 4 models in the Groq provider, with the model enums: LLAMA_4_MAVERICK_17B, LLAMA_4_SCOUT_17B.

  • We have added threadId to email metadata, for tracking the original email thread identifier.

Bugs Fixed

  • GPLA-4103: Unable to lookup content by original URI

  • GPLA-4109: Not parsing out hyperlinks from HTML list items

Last updated