●Change8

v0.14.13

Breaking Changes
📦 llamaindexView on GitHub →
⚠ 1 breaking✨ 19 features🐛 7 fixes⚡ 1 deprecations🔧 17 symbols

Summary

This release introduces distributed data ingestion via Ray, token-based code splitting, and new integrations like Apertis LLM and Volcengine MySQL vector store. Notably, ChatMemoryBuffer has been replaced by the generic Memory class.

⚠️ Breaking Changes

  • The `ChatMemoryBuffer` class has been replaced by the generic `Memory` class in `llama-index-core`. Users relying on `ChatMemoryBuffer` must update their code to use `Memory` instead.

Migration Steps

  1. Replace usage of `ChatMemoryBuffer` with `Memory` in `llama-index-core`.
  2. When using `llama-index-vector-stores-milvus`, update calls to add/delete operations to use `milvus_partition_name` instead of the previous partition parameter.

✨ New Features

  • Added `early_stopping_method` parameter to agent workflows in `llama-index-core`.
  • Added token-based code splitting support to `CodeSplitter` in `llama-index-core`.
  • Added `RayIngestionPipeline` integration for distributed data ingestion.
  • Added multi-modal version of the Condensed Conversation & Context.
  • Added configurable empty response message to synthesizers in `llama-index-core`.
  • Enabled use of ARNs for Bedrock Embedding Models in `llama-index-embeddings-bedrock`.
  • Improved Ollama batch embedding in `llama-index-embeddings-ollama`.
  • Added voyage-4 models support in `llama-index-embeddings-voyageai`.
  • Added `Apertis LLM` integration.
  • Enhanced structured predict methods for Anthropic in `llama-index-llms-anthropic`.
  • Added OpenRouter provider routing support in `llama-index-llms-openrouter`.
  • Added HuggingFace datasets reader integration (`llama-index-readers-datasets`).
  • Patentsview reader updated for API changes.
  • Revamped `YouRetriever` integration.
  • Added Parallel Web System tools.
  • Added Alibaba Cloud MySQL vector store integration.
  • Added Milvus partition name support, changing the partition parameter to `milvus_partition_name` in add/delete operations in `llama-index-vector-stores-milvus`.
  • Added hybrid search support to `llama-index-vector-stores-vertexaivectorsearch`.
  • Added Volcengine MySQL vector store integration.

🐛 Bug Fixes

  • Fixed `mean_agg` in `llama-index-core` to raise a `ValueError` when the input list is empty instead of returning a float.
  • Fixed `ReActChatFormatter` classmethod to use `cls` instead of the class name in `llama-index-core`.
  • Fixed preserving `input_tokens` in Anthropic stream_chat responses in `llama-index-llms-anthropic`.
  • Sanitized OpenAI structured output JSON schema name for generic Pydantic models in `llama-index-llms-openai`.
  • Fixed MongoDB async integration in `llama-index-vector-stores-mongodb`.
  • Fixed handling missing metadata for neo4j vector store in `llama-index-vector-stores-neo4jvector`.
  • Added `close` and `aclose` methods to the vector client in `llama-index-vector-stores-opensearch`.

Affected Symbols

ChatMemoryBufferMemoryCodeSplittermean_aggReActChatFormatterBedrock Embedding ModelsOllamaVoyageAI modelsRayIngestionPipelineAnthropic structured predict methodsChatMessage (in bedrock-converse)OpenAI structured output JSON schemaYouRetrieverMilvus add/delete operationsmongodb async integrationneo4j vector store metadata handlingOpenSearch vector client

⚡ Deprecations

  • The package `llama-index-llms-gemini` is deprecated as of this release.