Release Notes

llama-index-core [0.12.38]

• feat: Adding a very simple implementation of an embeddings cache (#18864)
• feat: Add cols_retrievers in NLSQLRetriever (#18843)
• feat: Add row, col, and table retrievers as args in NLSQLTableQueryEngine (#18874)
• feat: add configurable allow_parallel_tool_calls to FunctionAgent (#18829)
• feat: Allow ctx in BaseToolSpec functions, other ctx + tool calling overhauls (#18783)
• feat: Optimize get_biggest_prompt for readability and efficiency (#18808)
• fix: prevent DoS attacks in JSONReader (#18877)
• fix: SelectionOutputParser when LLM chooses no choices (#18886)
• fix: resuming AgentWorkflow from ctx during hitl (#18844)
• fix: context serialization during AgentWorkflow runs (#18866)
• fix: Throw error if content block resolve methods yield empty bytes (#18819)
• fix: Reduce issues when parsing "Thought/Action/Action Input" ReActAgent completions (#18818)
• fix: Strip code block backticks from QueryFusionRetriever llm response (#18825)
• fix: Fix get_function_tool in function_program.py when schema doesn't have "title" key (#18796)

llama-index-agent-azure-foundry [0.1.0]

• feat: add azure foundry agent integration (#18772)

llama-index-agent-llm-compiler [0.3.1]

• feat: llm-compiler support stream_step/astream_step (#18809)

llama-index-embeddings-google-genai [0.2.0]

• feat: add gemini embeddings tests and retry configs (#18846)

llama-index-embeddings-openai-like [0.1.1]

• fix: Pass http_client & async_http_client to parent for OpenAILikeEmbedding (#18881)

llama-index-embeddings-voyageai [0.3.6]

• feat: Introducing voyage-3.5 models (#18793)

llama-index-indices-managed-llama-cloud [0.7.1]

• feat: add client support for search_filters_inference_schema (#18867)
• feat: add async methods and blank index creation (#18859)

llama-index-llms-anthropic [0.6.19]

• feat: update for claude 4 support in Anthropic LLM (#18817)
• fix: thinking + tool calls in anthropic (#18834)
• fix: check thinking is non-null in anthropic messages (#18838)
• fix: update/fix claude-4 support (#18820)

llama-index-llms-bedrock-converse [0.6.0]

• feat: add-claude4-model-support (#18827)
• fix: fixing DocumentBlock usage within Bedrock Converse (#18791)
• fix: calling tools with empty arguments (#18786)

llama-index-llms-cleanlab [0.5.0]

• feat: Update package name and models (#18483)

llama-index-llms-featherlessai [0.1.0]

• feat: featherless-llm-integration (#18778)

llama-index-llms-google-genai [0.1.14]

• fix: Google GenAI token counting behavior, add basic retry mechanism (#18876)

llama-index-llms-ollama [0.5.6]

• feat: Attempt to automatically set context window in ollama (#18822)
• feat: use default temp in ollama models (#18815)

llama-index-llms-openai [0.3.44]

• feat: Adding new OpenAI responses features (image gen, mcp call, code interpreter) (#18810)
• fix: Update OpenAI response type imports for latest openai library compatibility (#18824)
• fix: Skip tool description length check in OpenAI agent (#18790)

llama-index-llms-servam [0.1.1]

• feat: add Servam AI LLM integration with OpenAI-like interface (#18841)

llama-index-observability-otel [0.1.0]

• feat: OpenTelemetry integration for observability (#18744)

llama-index-packs-raptor [0.3.2]

• Use global llama_index tokenizer in Raptor clustering (#18802)

llama-index-postprocessor-rankllm-rerank [0.5.0]

• feat: use latest rank-llm sdk (#18831)

llama-index-readers-azstorage-blob [0.3.1]

• fix: Metadata and filename in azstorageblobreader (#18816)

llama-index-readers-file [0.4.8]

• fix: reading pptx files from remote fs (#18862)

llama-index-storage-kvstore-postgres [0.3.1]

• feat: Create PostgresKVStore from existing SQLAlchemy Engine (#18798)

llama-index-tools-brightdata [0.1.0]

• feat: brightdata integration (#18690)

llama-index-tools-google [0.3.1]

• fix: GmailToolSpec.load_data() calls search with missing args (#18832)

llama-index-tools-mcp [0.2.2]

• feat: enhance SSE endpoint detection for broader compatibility (#18868)
• feat: overhaul BasicMCPClient to support all MCP features (#18833)
• fix: McpToolSpec fetch all tools given the empty allowed_tools list (#18879)
• fix: add missing BasicMCPClient.with_oauth() kwargs (#18845)

llama-index-tools-valyu [0.2.0]

• feat: Update to valyu 2.0.0 (#18861)

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

• feat: Add Vector Index Compression support for Azure Cosmos DB Mongo vector store (#18850)

llama-index-vector-stores-opensearch [0.5.5]

• feat: add filter support to check if a metadata key doesn't exist (#18851)
• fix: dont pass in both extra_info and metadata in vector store nodes (#18805)