v0.13.2
📦 llamaindexView on GitHub →
✨ 6 features🐛 9 fixes🔧 9 symbols
Summary
This release introduces the Superlinked retriever integration, enhances PowerPoint data extraction, and updates ClickHouse for new vector search capabilities, alongside various bug fixes for Anthropic, AzureOpenAI, and Ollama integrations.
✨ New Features
- Added ability to disable streaming in agents within llama-index-core.
- Enhanced PowerPoint reader with comprehensive content extraction in llama-index-readers-file.
- Added Superlinked retriever integration.
- Added support for openai-oss models to BedrockConverse.
- Added handlers for custom types and pydantic models in MCP tools.
- Updated ClickHouse integration to support new native vector search capabilities.
🐛 Bug Fixes
- Fixed NLTK_DATA environment variable respect in llama-index-core.
- Fixed order preservation and fetching in batch non-cached embeddings for a/get_text_embedding_batch().
- Fixed embedding output access and normalization in Ollama embeddings.
- Fixed Anthropic extended thinking error when used with tools.
- Fixed role response detection in Ollama LLM.
- Fixed AzureOpenAI streaming token usage reporting.
- Fixed persistence and loading issues in BM25 retriever.
- Fixed special character normalization in PostgreSQL ts_query.
- Fixed and updated Cognee integration.
🔧 Affected Symbols
a/get_text_embedding_batchNLTK_DATABedrockConverseAzureOpenAIPptxReaderBM25RetrieverClickHouseVectorStorePGVectorStoreSuperlinkedRetriever