●Change8

v0.13.3

📦 llamaindexView on GitHub →
✨ 4 features🐛 10 fixes⚡ 1 deprecations🔧 12 symbols

Summary

This release introduces HerokuEmbeddings support, Qdrant sharding, and GPT-5 model support, alongside critical stability fixes for streaming responses, empty vector stores, and concurrent reader usage.

Migration Steps

  1. Update llama-index-llms-upstage and migrate away from the deprecated solar-pro model.

✨ New Features

  • Added support for HerokuEmbeddings via llama-index-embeddings-heroku.
  • Enhanced OllamaEmbedding with instruction support.
  • Added support for gpt-5-chat-latest model in OpenAI LLM.
  • Added support for sharding in Qdrant vector store.

🐛 Bug Fixes

  • Added timeouts on image .get() requests in core.
  • Fixed message loss bug in StreamingAgentChatResponse.
  • Fixed crash when retrieving from an empty vector store index.
  • Fixed ContextChatEngine to correctly handle QueryBundle inputs.
  • Fixed faithfulness evaluator crash when images are missing.
  • Fixed prompt caching with CachePoint in Anthropic LLM.
  • Fixed SageMakerEndpoint constructor to correctly handle region_name and kwargs.
  • Fixed ConfluenceReader to support concurrent use of multiple instances.
  • Fixed Chroma vector store get_nodes() behavior with empty node IDs.
  • Resolved AttributeError in TencentVectorDB CollectionParams.filter_fields access.

🔧 Affected Symbols

StreamingAgentChatResponseVectorStoreIndexContextChatEngineQueryBundleHerokuEmbeddingsOllamaEmbeddingAnthropicSageMakerEndpointConfluenceReaderChromaVectorStore.get_nodesQdrantVectorStoreTencentVectorDB.CollectionParams

⚡ Deprecations

  • The solar-pro model in llama-index-llms-upstage has been removed.