v0.14.24

🐛 Bug Fixes

• Close HotpotQA evaluator file handles.
• Preserve oversized leaf nodes in CodeSplitter.
• Preserve multiblock chat history writes.
• Skip empty nodes in HTMLNodeParser.
• Respect mmr_threshold=0 in MMR embedding search.
• Keep all nodes per document in IngestionPipeline upserts.
• Do not mark *args/**kwargs as required tool parameters.
• Default MetadataFilters condition to AND when None.
• Honor agent structured_output_fn/output_cls inside AgentWorkflow.
• Use word tokenization for stopword removal in SemanticDoubleMergingSplitterNodeParser.
• Restore prompt helper fallback to supplied LLM metadata.
• Populate response text on streaming chat response when writing to memory.
• Clarify FactExtractionMemoryBlock condense prompt to return a full snapshot.
• Fix a few typos in llama-index-core and llama-index-node-parser-slide.
• Give each CitationQueryEngine citation node its own id and offsets.
• Persist SimpleChatStore without escaping non-ascii.
• Make aioboto3 optional in llama-index-embeddings-bedrock.
• Validate non-empty Bedrock embedding payload before AWS call in llama-index-embeddings-bedrock.
• Set chunk_size=32000 and fix input format for contextualizedembeddings in llama-index-embeddings-voyageai.
• Correct schema, vector ranking, filters and batching in property graph store for falkordb.
• Send query_history in retriever payload for dashscope.
• Repair image ingestion crash, embedding_model typo, and linting issues in lancedb.
• Fix Sonnet 5 function calling for anthropic LLM.
• Correct Opus 4.6 context window to 1M in llama-index-llms-anthropic.
• Make aioboto3 optional in llama-index-llms-bedrock-converse.
• Do not overwrite thought signatures in llama-index-llms-google-genai.
• Do not send default params in gemini genconfig in llama-index-llms-google-genai.
• Report the model's effective context window in llama-index-llms-llama-cpp.
• Add GPT-5.6 models to supported OpenAI models.
• Bump instrumentation version in otel integration to 0.5.0 as minimum.
• Improve otel span ID handling.
• Raise ValueError instead of fabricating tool_call_id in ag-ui.
• Persist AG-UI frontend tool messages.
• Close file handles and http connections in reader integrations for alibabacloud-aisearch and huggingface-fs.
• Add telemetry to couchbase reader.
• Bound HWP section decompression in file and hwp readers.
• Fix GCS reader README import paths.
• Add asyncio_mode=auto to unblock async tests in CI for github-reader.
• GoogleDriveReader returns None instead of a list when an error is swallowed.
• Fix MemosReader default endpoint and metadata.