●Change8

v0.12.45

Breaking Changes
📦 llamaindexView on GitHub →
⚠ 1 breaking✨ 7 features🐛 5 fixes⚡ 2 deprecations🔧 9 symbols

Summary

This release introduces content block support for tools, adds new AWS Claude models, and updates the core memory and UI event standards while fixing several parser and vector store bugs.

⚠️ Breaking Changes

  • The base index now uses 'id_' instead of 'get_doc_id()', which may affect custom index implementations relying on the old method name.

Migration Steps

  1. Update custom index implementations to use 'id_' instead of 'get_doc_id()'.
  2. Migrate away from deprecated agent packages to supported core alternatives.
  3. Ensure aioboto3 is updated if using Bedrock-related integrations.

✨ New Features

  • Tools can now output content blocks.
  • Added chat UI events and models to the core package.
  • Added new AWS Claude models to Bedrock LLM.
  • Added dimensions parameter to AzureOpenAIEmbedding.
  • Updated aioboto3 dependency for Bedrock embeddings and LLMs.
  • Adapted Mem0 memory to the new framework memory standard.
  • Added proper async support to Google Search tool spec.

🐛 Bug Fixes

  • Fixed Node loading from ingestion cache.
  • Fixed SemanticDoubleMergingSplitterNodeParser not respecting max_chunk_size.
  • Fixed DashScope Qwen Assistant API error response and tool_calls extraction.
  • Cleaned up results in GoogleSearchToolSpec.
  • Fixed sparse query issues in Postgres vector store.

🔧 Affected Symbols

BaseIndex.get_doc_idBaseIndex.id_SemanticDoubleMergingSplitterNodeParserAzureOpenAIEmbeddingGoogleSearchToolSpecPGVectorStoreDashScopeMem0MemoryContext

⚡ Deprecations

  • Older agent packages have been deprecated.
  • Usage and references to the deprecated Context get/set API have been removed.