LlamaIndex v0.14.24
LlamaIndex v0.14.24 updates core agent workflows, memory handling, and retrieval components while expanding integrations across major LLM providers including Anthropic, Google GenAI, Bedrock, and OpenAI.
原始内容为英文；当前页面提供中文导航与来源说明，具体事实请以原文为准。
为什么值得关注
LlamaIndex release v0.14.24 delivers broad updates across core and extension packages. Core Highlights: - Agent & Tooling: Ensures structured output functions/classes are respected inside AgentWorkflow, avoids marking args/ kwargs as required tool parameters, and properly populates response text during streaming chat when saving to memory. - Memory & Storage: Supports any AsyncDBChatStore in Memory, clarifies FactExtractionMemoryBlock condensing prompt, preserves multiblock chat history writes, and persists SimpleChatStore without escaping non-ASCII characters. - Retrieval & Processing: Implements async support for LLMRerank, fixes MMR embedding search with threshold=0, defaults MetadataFilters condition to AND when None, and preserves oversized leaf nodes in CodeSplitter. Ecosystem & Integration Updates: - Anthropic & Bedrock Converse: Adds support for Claude Sonnet 5, updates Opus 4.6 context window to 1M, and adds Claude Opus 5 to model allowlists. Bedrock integrations make aioboto3 optional and add support for disabling thinking types. - Google GenAI: Switches the library and docs default model to Gemini 3.7 Flash, avoids sending default generation config parameters, and prevents overwriting thought signatures. - OpenAI: Adds GPT-5.6 models to supported model lists. - Providers & Protocols: Adjusts VoyageAI contextualized embeddings chunk size (32000), fixes schema and vector ranking in FalkorDB property graph store, fixes LanceDB image ingestion crashes, and updates AG-UI protocol to raise a ValueError instead of generating artificial tool call IDs.
可执行摘要
LlamaIndex v0.14.24 introduces core fixes such as honoring structured outputs in AgentWorkflow, omitting args/ kwargs from required tool parameters, async LLMRerank support, and streaming response text population in memory. It also updates integrations for Bedrock, Anthropic, Google GenAI (defaulting to Gemini 3.7 Flash), and OpenAI.
- Agent 实用度
- 82/100
- 可信度
- 96%
- 机器格式
- JSON + Markdown
开发者应核对什么
- Upgrade llama-index-core and specific integration packages (e.g., llama-index-llms-anthropic, llama-index-llms-google-genai) to incorporate recent model support and bug fixes.
- Review tool signature definitions to ensure removal of explicit workarounds for args and kwargs parameter schemas.
- Audit AgentWorkflow structured output configurations to ensure output functions/schemas align with the now-enforced AgentWorkflow behavior.