Agno v3.0.11
Agno v3.0.11 introduces a unified knowledge retrieval pipeline with MMR and recency rerankers, deprecating vector DB-specific rerankers. It also adds structured cancellation stages, Y-API support, and multiple fixes across MCP, WebSockets, and data readers.
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
Agno v3.0.11 delivers a significant overhaul to its retrieval and execution systems. Key updates include: - Knowledge Retrieval Pipeline: Knowledge.search() now runs a knowledge-level retrieval pipeline that widens the candidate pool prior to reranking. Consequently, the reranker parameter on individual vector databases is deprecated in favor of Knowledge -level rerankers. - New Rerankers: Added MMRReranker for marginal relevance/diversity balancing, and RecencyReranker (focused on pgvector) using exponential timestamp decay. - Page Source Relocation & Typed Tools: Added migration inspection and dry-run capable migration for indexed documentation sources. Added typed execution results ( PageCommandResult , PageFileSystem.run command result ) and ranked SearchResult structures. - Run Metadata: Machine-readable cancellation stage field ( PENDING , EXECUTING , INTERRUPTED , or unknown ) added across RunOutput , TeamRunOutput , WorkflowRunOutput , and API schemas. - Integrations & LLM Support: Added YAPI provider, recognized DeepSeek thinking-mode IDs under OpenAILike , decoupled OpenAI response storage ( store=True ) from forced ID chaining via use previous response id , and updated AG-UI to support ag-ui-protocol 1.0 . - Critical Fixes: Sync hooks returning function call now execute correctly in async flows, MCP tool discovery over raw ClientSession now paginates through all pages, and WebSocket workflow submission rules now align with HTTP admission (pinning, session checks).
可执行摘要
Agno v3.0.11 decouples rerankers from specific vector databases to Knowledge.search(), introducing MMRReranker and RecencyReranker. It establishes machine-readable cancellation stages in run outputs, adds YAPI integration, patches async tool hook execution, fixes MCP pagination, and enforces UTF-8 across file operations.
- Agent 实用度
- 82/100
- 可信度
- 96%
- 机器格式
- JSON + Markdown
开发者应核对什么
- Migrate existing vector DB-level reranker configurations to the Knowledge -level reranker parameter.
- Update client-side cancellation handling to inspect the structured cancellation stage attribute instead of parsing error message strings.
- Review MCP integration setups using raw ClientSession to ensure compatibility with full pagination of discovered tools.