Version 1.1.0

Lumen AI 1.1 builds on the stable foundation of 1.0 with first-class LLM tool support, stronger retrieval-augmented generation, and a broad sweep of usability and reliability improvements.

LLM Tool Calls & MCP Support

• First-class support for LLM tool calls

  • New FunctionTool and MCPTool abstractions

• Expanded provider support

  • Dedicated Groq provider
  • Fixed non-streaming Anthropic response handling
  • Fixed annotation support for Google Gemini
  • Graceful multimodal content handling for non-vision providers
  • Cleaner LLM inheritance and provider configuration
  • Default mini model updated to GPT-5.4-mini

Retrieval-Augmented Generation

• ChromaDB as a vector store backend

  • New ChromaDBVectorStore with native embedding model support
  • Shared BaseOpenAIEmbeddings base class with normalized whitespace handling

• Smarter metadata handling

  • Per-exploration metadata vector stores for up-to-date context
  • Metadata better surfaced in agent prompts
  • Race condition in MetadataLookup indexing resolved

• New DocumentSummarizerAgent

  • Document ingestion and summarization as a first-class workflow

Upload, Export & Editors

• Improved upload experience

  • Configurable file size limits and FileDropper options
  • Simplified drag-and-drop flow with better guidance for new users
  • Deduplicated source references for multi-file uploads

• New export formats

  • SQLEditor: JSON and Markdown
  • VegaLiteEditor: WebP, TIFF, and EPS
  • Copy button tooltips now reflect the active editor language

UX & Interface

• More flexible explorations

  • Explorations can now be created without requiring a data pipeline
  • Direct editing of chat messages via edit_callback
  • Label lineage and distinct follow-ups improve conversation traceability

• Refined controls and layout

  • Improved controls abstraction, report UX, and pop-out insertion logic
  • Exploration "prevent and allow" limits for finer administrative control
  • Various prompt improvements for better SQL, revision, and insertion quality

Data & SQL

• Broader SQL capabilities

  • BETWEEN query support via slice syntax in SQLFilter
  • Fixed transforms on unselected data and list-typed source.tables

• Better table rendering

  • Nested JSON objects flattened correctly, preventing [object Object] display
  • Fixed horizontal overflow with long string columns

Stability

• Pipeline and source fixes
• Broader reliability improvements

Documentation

• Updated LLM docs and added the AI Catalyst guide
• Documented template_overrides with subclassing patterns
• Improved contributing.md formatting and fixed a broken README link

Contributors

Many thanks to @philippjfr and @ahuang11, and a special welcome to the four new contributors making their first commits to the project: @ghostiee-11, @SuMayaBee, @SanketMeghale, and @crypticsaiyan.

Full changelog: v1.0.0 → v1.1.0