Skip to main content

0.61.0

· 2 min read

Features

  • Introduce UI language support - Wren AI is now available in English, Traditional Chinese, Simplified Chinese, Japanese, Korean, Spanish and French, selected from your language preference
  • Introduce the redesigned modeling canvas - a new automatic layout with a layout picker remembered per project, plus sidebar search, hide/show and collapse for large models
  • Introduce Git access for Agentic Mode projects - bind a local directory with the wren CLI, then use ordinary git push and git pull to move models, views, relationships and knowledge between your machine and Wren AI
  • Agentic Mode is now generally available - the Beta label has been removed
  • Support encrypted private keys for Snowflake key-pair authentication - supply the passphrase alongside the key

Improvements

  • Add allowUserGuide and userGuidePrompt options to the public API - embedded and white-label callers can turn off Wren's built-in product help or supply their own guide
  • Improve chart quality and speed in Agentic Mode - answers lead with the written explanation, charts are checked for clipped or overlapping labels before they are shown, and chart data is queried directly instead of being retyped by the model
  • Improve Agentic Mode reliability - conversations no longer stall when the model connection drops mid-response, persistent memory and always-apply knowledge rules are honored consistently, and errors are reported in plain language
  • Render mathematical notation in answers - LaTeX expressions display correctly in agent replies and artifacts
  • Show when Interactive mode reused a saved question - the thinking steps now include the matched Question-SQL pair
  • Improve SQL accuracy for Spark-based sources - expanded Spark SQL function knowledge for Spark, Athena Spark and EMR Spark
  • Retire project-level MCP connections - the organization-level MCP endpoint is now the single connection point; per-project MCP URLs no longer work
  • Improve how Agentic Mode project knowledge is applied to answers through MCP, the public API, Teams and Slack