0.64.0

Features

• Introduce answer feedback through the public API and MCP - every answer now returns an answer ID, so you can rate it from your own application, and an MCP client can report a wrong answer in the conversation; each rating is attached to that answer's trace
• Introduce project descriptions for AI assistants - describe what each project covers and what it does not in Project Settings or through the API, and assistants connected over MCP use it to pick the right project, or ask you when more than one could answer; the Wren agent can draft the description for you
• Review what the agent publishes in Agentic Mode - knowledge, model changes and skills the agent drafts now wait for your decision on a confirmation card, with a preview of the change, before anything is published
• Support Oracle 12c - queries run correctly on Oracle 12c databases, including national-character text and timestamps with time zones

Improvements

• Improve the knowledge the agent writes - entries follow one structure with a name and usage notes, references to your models are checked before an entry is saved, and the knowledge templates follow the same format
• Show when a long Agentic conversation is condensing earlier messages - the step appears among the thinking steps with its duration instead of a silent wait
• Improve the agent's response time in Agentic Mode - the first output arrives sooner on new threads and on follow-up questions
• Show who asked each question in thread tracing - traces from the web app and from MCP now carry the user
• Record organization API key changes in the activity log - creating, renaming and deleting a key each leave an entry
• Improve schema change detection for Agentic projects - models keep their source table name after an agent sync or an MDL import, so they are no longer reported as deleted