Skip to main content

Release 7.8.0 - ConvyAI

What's new​

  • Xference AI Model provider - Xference is now available as a native AI Model provider, with inference running on Italian datacenters. Configure it with an endpoint and API key, choose from the exposed completion models (Qwen 3.6, Gemma 4 3B, Gemma 4 12B) or enter a custom one, and use the BAAI/bge-m3 embedding model. See Xference for details.

  • Xference in AI Agent nodes - AI Model of type Xference can now be selected in the AI Agent node of both the FlowDesigner and the Agentic Designer, with full support for MCP Server tools, JavaScript functions, and image analysis, on par with the other providers.

  • Xference for Knowledge Base indexing and retrieval - Xference can now be configured as an embedding provider for Knowledge Base indexing batches, and an Xference AI Model can be used in the Knowledge Base node to query indexes built with the BAAI/bge-m3 embedding model. See Indexing for details.

Improvements​

  • Outbound to WhatsApp BSUID contacts - Building on the BSUID recognition introduced in 7.7.0, ConvyAI can now also target WhatsApp users identified only by BSUID as outbound recipients. Map a BSUID column when importing a contact list, select or type a BSUID when building a massive campaign or a one-to-one outbound from Agent Desktop - the message is delivered through the correct identifier, phone number when available and BSUID otherwise. See Contact Lists, Campaign Lists, and Agent Desktop Outbound for details.