Agno v3.0.7
Agno v3.0.7 introduces atomic page storage, public serving capabilities for Agents and Teams, fixes for provider crashes (Meta Llama, AIMLAPI, Anthropic tool calls), and breaking changes to the Knowledge constructor and PgVector search settings.
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 v3.0.7 delivers several architectural enhancements and stability fixes: - Page Storage & Public Serving : Introduces agno.knowledge.page with atomic metadata/vector publication and transaction-local planner controls. Adds PublicSurface for serving selected Agents, Teams, native stateless MCP, and authenticated durable-sync Workflows with quotas, admission limits, and internal auth. A bounded read-only PageFileSystem toolkit is also introduced. - Agent & Workflow Enhancements : Agent dependency callables can now receive run input and session context. Failed workflow steps are explicitly rendered as StepError , and workflow run state finalization now completes prior to yielding terminal streaming failure events. - Critical Fixes : Resolves TypeError crashes in Meta Llama and AIMLAPI format message handlers due to mismatched argument handling. Fixes Anthropic tool serialization to properly send empty object arguments "{}" . Corrects a memory leak on Python 3.13 where cached tool wrappers retained caller frames, Agents, and sessions. Anonymous native agent failures now properly return 503 instead of exposing diagnostics in a 200 response. - Breaking Changes : Knowledge constructor arguments are now strictly keyword-only (e.g., Knowledge(name="docs") ). content db is preferred over contents db . Page search now respects configured PgVector HNSW ef search values rather than forcing 200.
可执行摘要
Agno v3.0.7 adds agno.knowledge.page storage, PublicSurface endpoint serving for Agents/Teams/MCP, and PageFileSystem . It introduces breaking changes by making Knowledge constructor arguments keyword-only and respecting configured PgVector ef search . Key fixes include resolving a Python 3.13 tool wrapper memory leak, handling Anthropic empty tool arguments, and fixing format message signature mismatches for Meta…
- Agent 实用度
- 82/100
- 可信度
- 96%
- 机器格式
- JSON + Markdown
开发者应核对什么
- Update all Knowledge(...) constructor calls to use keyword-only arguments (e.g., Knowledge(name="docs") ).
- Explicitly set ef search=200 in PgVector search configurations if your application relied on the previously hardcoded default.
- Rebuild interim page indexes that defaulted to ef construction=64 if using main branch builds.
- Install the new agno pages extra if using the new page storage and filesystem tools.