●Change8

v0.12.47

Breaking Changes
📦 llamaindexView on GitHub →
⚠ 1 breaking✨ 6 features🐛 3 fixes⚡ 1 deprecations🔧 8 symbols

Summary

This release unifies Multi Modal LLMs into the base LLM class, introduces a security fix for the RAG CLI, and adds several integrations including LanceDB MultiModal and Anthropic citations.

⚠️ Breaking Changes

  • Multi Modal LLMs are no longer separate classes; they now extend the base LLM class. Existing multi-modal specific implementations may need to be updated to use the unified LLM interface.

Migration Steps

  1. Update Multi Modal LLM implementations to use the base LLM class and ImageBlock for multi-modal features.
  2. Review agent .run() calls to ensure the new default max_iterations of 20 is appropriate for your use case.

✨ New Features

  • Added default max_iterations=20 to agent .run() method.
  • Set tool_required=True for FunctionCallingProgram and structured LLMs where supported.
  • Added Anthropic citations support.
  • Added support for Image prompts in OCI Generative AI Llama models.
  • New integration for LanceDB MultiModal AI LakeHouse.
  • Added Firecrawl integration source to web readers.

🐛 Bug Fixes

  • Fixed missing raw data in agent workflow events.
  • Fixed parsing of empty lists in JSON output parsing.
  • Fixed command injection vulnerability in RAG CLI --clear flag.

🔧 Affected Symbols

LLMMultiModalLLMImageBlockFunctionCallingProgramAgentRunner.runLanceDBIndexAnthropicOCIGenAI

⚡ Deprecations

  • Deprecated Multi Modal LLM specific classes in favor of base LLM classes with ImageBlock support.