Skip to content

v2.6.1

Choose a tag to compare

View all tags
@jeffwu-1999 jeffwu-1999 released this 22 Sep 01:41
· 63 commits to main since this release
v2.6.1
02a54cc
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

🚀 Nexent：开源智能体平台 / Nexent: Open Source Intelligent Agent Platform

我们很高兴地宣布Nexent v2.6.1 正式发布！🎉

Nexent 是一个开源智能体平台，能够将流程的自然语言转化为完整的多模态智能体 —— 无需编排，无需复杂拖拉拽。基于 MCP 工具生态，Nexent 提供强大的模型集成、数据处理、知识库管理、零代码智能体开发能力。我们的目标很简单：将数据、模型和工具整合到一个智能中心中，使日常工作流程更智能、更互联。

We are excited to announce that Nexent v2.6.1 is released! 🎉

Nexent is an open-source agent platform that turns process-level natural language into complete multimodal agents — no diagrams, no wiring. Built on the MCP tool ecosystem, Nexent provides model integration, data processing, knowledge-base management, and zero-code agent development. Our goal is simple: to bring data, models, and tools together in one smart hub, making daily workflows smarter and more connected.

重构优化 / Refactoring & Improvements

  • 重构优化人在回路（human-in-the-loop）交互实现：一次 query 对应一次独立 Agent 运行，需要人工输入时结束本轮并释放资源，用户输入后开启新一轮，任务通过会话历史延续，降低长时等待与资源占用。 / Refactored the human-in-the-loop interaction model so that one query maps to one independent agent run: the run ends and releases resources when human input is needed, and the user's reply starts the next run, with the task continuing through conversation history — reducing long waits and resource usage.
  • 优化主动反问与用户打断流程：支持单选、多选、文本三类澄清问题，用户在 Agent 运行中给出的意见可作为新一轮输入继续执行。 / Improved agent-initiated clarification and user-interruption flows with single-choice, multiple-choice, and free-text questions, where user guidance is submitted as the next run's input.
  • 优化 CodeAgent 输出协议：要求显式终止信号，区分 code_action 与 final_answer_envelope，并兼容带 reasoning 前缀的代码动作，减少静默中断。 / Improved the CodeAgent output protocol to require an explicit termination signal, distinguish code_action from final_answer_envelope, and accept reasoning-prefixed code actions, reducing silent stalls.
  • 完善模型错误处理：统一稳定的错误分类（限流耗尽、服务不可用、超时、连接错误、鉴权错误、上下文溢出等）并在 Agent 流边界暴露；瞬时错误改为静默重试，同时支持 OpenAI 兼容客户端的 httpx2 超时配置。 / Improved model error handling with stable error categories (rate limit exhausted, service unavailable, timeout, connection error, authentication error, context overflow) surfaced consistently at the agent stream boundary, plus silent retry for transient failures and httpx2 timeout support for OpenAI-compatible clients.
  • 优化 AIDP 知识库：完善文档状态跟踪（上传中/处理中/提取中/已完成/失败）、知识库与通道查询、文档历史读取，并修复保存与检索相关问题。 / Improved the AIDP knowledge base with document status tracking (uploading / processing / extracting / completed / failed), knowledge-base and channel queries, and document history retrieval, and fixed issues in saving and retrieval.
  • 优化评估（evaluation）试运行链路：经由带鉴权的 Config-to-Runtime 代理执行，Agent 执行与评分统一留在 Runtime 侧。 / Improved the evaluation trial flow to run through the authenticated Config-to-Runtime proxy, keeping agent execution and evaluator scoring on the Runtime side.
  • 优化沙箱容器权限配置，容器用户可以正常创建文件夹与文件。 / Improved sandbox container permission settings so the container user can create folders and files normally.

Bug 修复 / Bug Fixes

  • 修复编辑 ModelEngine 模型时 ssl_verify 被错误置为 True 导致连通性中断的问题，编辑与批量编辑路径均改为基于载荷 URL 判断（兼容自签名证书）。 / Fixed ModelEngine model edits incorrectly flipping ssl_verify to True and silently breaking connectivity; both edit and batch-edit paths now judge by payload URL with a fallback to the stored record (self-signed certificates supported).
  • 修复 ModelEngine provider 列表请求未分发到专用端点的问题。 / Fixed ModelEngine provider listing not being dispatched to the dedicated endpoint.
  • 修复自定义参数删除不持久（null 标记丢失）、参数被模型默认值覆盖、以及按 Agent 覆盖容量不生效的问题。 / Fixed custom parameter deletion not persisting (missing null markers), parameters reappearing from model defaults, and per-agent capacity overrides not taking effect.
  • 修复 ModelRequest.model_id 字段被展开进 INSERT 列清单从而注入空主键、导致模型创建失败的问题。 / Fixed ModelRequest.model_id being spread into INSERT column lists, which injected a null primary key and broke model creation.
  • 修复 Agent 未按预期显式终止以及静默恢复路径的问题。 / Fixed agents not terminating explicitly as expected and issues in the silent-recovery path.
  • 修复人在回路表单投递、事件顺序与调度并发相关的多个缺陷。 / Fixed several defects around human-in-the-loop form delivery, event ordering, and scheduler concurrency.

文档与部署变化 / Docs & Deployment Changes

  • 更新数据库迁移 v2.6.1_001_remove_human_interaction.sql：在确认相关进程全部停止后清理退役交互引擎的表结构与函数，普通会话消息与消息单元保持不变。 / Updated migration v2.6.1_001_remove_human_interaction.sql to clean up the retired interaction engine's tables and function after all related processes have stopped; ordinary conversation messages and units are preserved.
  • 同步更新 Docker Compose（含生产编排）、K8s Helm values/configmap/secrets、部署脚本与环境变量模板。 / Updated Docker Compose (including production orchestration), K8s Helm values/configmap/secrets, deployment scripts, and env templates.
  • 更新北向 API 与开始对话使用文档，保持与新交互流程一致。 / Updated the northbound API and start-chat guides to match the new interaction flow.

What's Changed

Full Changelog: v2.6.0...v2.6.1

Contributors

jeffwu-1999
Assets 2
Loading