●Change8

v0.12.42

📦 llamaindex
✨ 8 features🐛 8 fixes⚡ 1 deprecations🔧 8 symbols

Summary

This release introduces support for OpenAI's O3-pro and MistralAI reasoning models, adds a new multi-modal OpenAI-like LLM, and provides various fixes for async memory operations and figure retrieval.

Migration Steps

  1. Update llama-index-postprocessor-bedrock-rerank to use BedrockRerank class name.
  2. Ensure aioboto3 is updated if using Bedrock embeddings.

✨ New Features

  • Added reasoning support to MistralAI LLM and Magistral.
  • Added Day 0 support for OpenAI O3-pro model.
  • Introduced OpenAI-like multi-modal LLM support.
  • Added ArtifactEditorToolSpec for editing Pydantic objects.
  • Integrated figure retrieval SDK in LlamaCloud managed indices.
  • Added ability to exclude source fields from OpenSearch query responses.
  • Added label truncation to workflow visualization.
  • Updated aioboto3 dependency in Bedrock embeddings.

🐛 Bug Fixes

  • Fixed memory operations to use async variants within async functions in core.
  • Fixed input message passing to memory get in core.
  • Switched from hashing to UUID in SQLTableNodeMapping for broader compatibility.
  • Fixed None type handling for raw_figure_nodes in LlamaCloud figure retrieval.
  • Skipped tool description length check in OpenAI response API.
  • Fixed filename hashing robustness in papers reader.
  • Fixed integration bugs in Perplexity LLM.
  • Corrected documentation and tool integration for ElevenLabs voice agents.

🔧 Affected Symbols

SQLTableNodeMappingBedrockRerankAWSBedrockRerankArtifactEditorToolSpecMistralAIOpenAIpage_figure_nodes_to_node_with_scoreOpenSearchVectorStore

⚡ Deprecations

  • Prefer 'BedrockRerank' over 'AWSBedrockRerank' in llama-index-postprocessor-bedrock-rerank.