Release Notes

[2025-08-14]

llama-index-core [0.13.2]

• feat: allow streaming to be disabled in agents (#19668)
• fix: respect the value of NLTK_DATA env var if present (#19664)
• fix: Order preservation and fetching in batch non-cached embeddings in a/get_text_embedding_batch() (#19536)

llama-index-embeddings-ollama [0.8.1]

• fix: Access embedding output (#19635)
• fix: use normalized embeddings (#19622)

llama-index-graph-rag-cognee [0.3.0]

• fix: Update and fix cognee integration (#19650)

llama-index-llms-anthropic [0.8.4]

• fix: Error in Anthropic extended thinking with tool use (#19642)
• chore: context window for claude 4 sonnet to 1 mln tokens (#19649)

llama-index-llms-bedrock-converse [0.8.2]

• feat: add openai-oss models to BedrockConverse (#19653)

llama-index-llms-ollama [0.7.1]

• fix: fix ollama role response detection (#19671)

llama-index-llms-openai [0.5.3]

• fix: AzureOpenAI streaming token usage (#19633)

llama-index-readers-file [0.5.1]

• feat: enhance PowerPoint reader with comprehensive content extraction (#19478)

llama-index-retrievers-bm25 [0.6.3]

• fix: fix persist+load for bm25 (#19657)

llama-index-retrievers-superlinked [0.1.0]

• feat: add Superlinked retriever integration (#19636)

llama-index-tools-mcp [0.4.0]

• feat: Handlers for custom types and pydantic models in tools (#19601)

llama-index-vector-stores-clickhouse [0.6.0]

• chore: Updates to ClickHouse integration based on new vector search capabilities in ClickHouse (#19647)

llama-index-vector-stores-postgres [0.6.3]

• fix: Add other special characters in ts_query normalization (#19637)