●Change8

v0.13.1

Breaking Changes
📦 llamaindexView on GitHub →
⚠ 1 breaking✨ 12 features🐛 8 fixes⚡ 1 deprecations🔧 11 symbols

Summary

This release introduces several new integrations including Heroku LLMs and Bedrock AgentCore tools, adds support for GPT-5 and Anthropic Opus 4.1, and deprecates the managed Llama Cloud index in favor of a new services package.

⚠️ Breaking Changes

  • The Ollama embedding integration now uses the /embed endpoint instead of /embeddings, which may require an updated Ollama server version.

Migration Steps

  1. Switch from llama-index-indices-managed-llama-cloud to the new llama-cloud-services package.
  2. Ensure Ollama server is updated to support the /embed endpoint if using llama-index-embeddings-ollama v0.8.0+.

✨ New Features

  • Enabled partially formatted system prompt for ReAct agent.
  • Added support for Voyage context embeddings.
  • Updated Kuzu graph store integration to the latest SDK.
  • Anthropic citation support moved to non-beta.
  • Added support for Anthropic Opus 4.1.
  • New Heroku LLM integration (llama-index-llms-heroku).
  • Added support for GPT-OSS NIM in NVIDIA integration.
  • Added support for GPT-5 in OpenAI integration.
  • Support for Presidio entities in postprocessor.
  • Added metadata filtering support to BM25 Retriever.
  • New Bedrock AgentCore browser and code interpreter toolspecs.
  • Added wait logic for critical operations in Baidu VectorDB.

🐛 Bug Fixes

  • Safer token counting in messages for core library.
  • Fixed Document truncation in FunctionTool._parse_tool_output.
  • Fixed isinstance check in OpenAI integration.
  • Fixed reasoning_effort parameter and added custom parameters for Upstage.
  • BM25 Retriever now allows top_k values greater than the number of nodes.
  • Fixed filter operators and added stores_text support for Baidu VectorDB.
  • Fixed special character bug in PGVectorStore queries.
  • Changed ts_query definition in Postgres to avoid double-stemming.

🔧 Affected Symbols

ReActAgentFunctionTool._parse_tool_outputOllamaEmbeddingKuzuGraphStoreAnthropicHerokuOpenAIBM25RetrieverPresidioPostprocessorPGVectorStoreBaiduVectorDB

⚡ Deprecations

  • llama-index-indices-managed-llama-cloud is deprecated in favor of llama-cloud-services.