v2.7.0

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

我们很高兴地宣布Nexent v2.7.0 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 v2.7.0 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.

---

新功能 / New Features

• 推出智能体工作台（Agent Workbench）：启动后直达工作台界面，支持以自然语言创建智能体，并提供 Skill 与知识库选择、创建历史、失败重试与恢复、会话续接等完整创建流程。 / Introduced the Agent Workbench, which lands directly on the workbench after launch and lets users create agents from natural language, with Skill and knowledge-base pickers, creation history, retry and recovery, and conversation resume.
• 内置官方智能体资产：随仓库提供政务（招标分析、公文写作、报告识别、安全报告、社媒洞察）与医疗（血检、病例生成、造影剂量、肺 CT 诊断/质控、医保审核）共 11 个预置智能体，含配套知识库与技能，并提供一键部署脚本。 / Bundled official agent assets: 11 pre-built agents across government (bid analysis, official document writing, report identification, security reporting, social media insight) and healthcare (blood test, case generation, contrast dosage, lung CT diagnosis/QC, insurance review), with companion knowledge bases, Skills, and a one-click deployment script.
• 支持安全审计日志：在 API Key、CAS、用户组、邀请、北向接口、OAuth、租户与用户管理等关键操作上记录审计流水。 / Added security audit logging across API key, CAS, group, invitation, northbound, OAuth, tenant, and user-management operations.
• 完善租户资源规格限额：对单租户的智能体、知识库、MCP、Skill、评测集与会话等资源设置上限，配套统一错误码、429 响应与中英文提示。 / Enforced tenant resource quotas for agents, knowledge bases, MCP services, Skills, evaluation sets, and conversations, with unified error codes, 429 responses, and localized messages.
• 重构模型配置页：以默认模型槽位网格替代原有弹窗，槽位带优先级徽标、连通性状态与一键校验，模型库保留表格、筛选与增删改流程。 / Redesigned the model configuration page with a default-model slot grid (priority badges, per-slot connectivity hints, one-click verification) while keeping the model library table, filters, and full CRUD flows.
• 支持模型调用日志分类与路由能力，便于按类别追踪与排查模型调用。 / Added classification and routing for model invocation logs to make model calls easier to track and troubleshoot.
• 支持将用户上下文传递至 MCP 工具与 A2A 智能体，实现工具侧鉴权。 / Passed user context through to MCP tools and A2A agents for tool-side authorization.
• 新增智能体发布与使用指引，帮助用户快速上手与发布智能体。 / Added an agent publishing and usage guide to help users get started and publish agents quickly.
• 支持模型思考挡位（reasoning effort）配置，并提供 MCP 地址与配置校验工具。 / Added model reasoning-effort configuration plus MCP URL and config validation utilities.
• 优化资源管理体验：统一资源页与资源卡片、优化超管管理页与 Skill 列表操作入口。 / Unified the resource page and resource cards, and polished the super-admin management page and Skill list actions.
• 优化智能体仓库与列表：支持分页列表中的仓库状态展示、卡片标签布局与名称溢出处理。 / Improved the agent repository and lists with repository status in paged views, card tag layout, and name-overflow handling.
• 前端构建提速：升级 Next.js 以启用更快的 Turbopack 构建。 / Upgraded Next.js for faster Turbopack builds.

Bug 修复 / Bug Fixes

• 修复租户管理员无法管理本租户模型的问题：/model/manage/* 改为「SU 任意租户、ADMIN 仅限本租户」的角色加租户作用域校验，恢复资源管理页模型列表与增删改。 / Fixed tenant admins being unable to manage their own tenant's models: /model/manage/* now uses a role-plus-tenant-scope check (SU for any tenant, ADMIN for its own), restoring the model list and CRUD on the resource page.
• 修复模型权限隔离：模型管理仅对管理员开放，DEV 角色不再看到模型页，同时保留只读权限供智能体编辑选用模型。 / Fixed model permission isolation: model management is restricted to admins and the model page is hidden from DEV, while read access remains for agent editors.
• 修复技能标签数据异常导致智能体页面崩溃的问题，并对标签做前后端双重规范化。 / Fixed an agents-page crash caused by malformed skill tags, with tag normalization on both frontend and backend.
• 修复配额分页与平台配额面板：知识库配额明细分页、面板加载态与公平分摊显示修正。 / Fixed quota breakdown pagination and the platform quota panel (KB pagination, loading states, and fair-share display).
• 修复监控异常上报，完善监控身份标识与指标归属。 / Fixed abnormal monitoring reporting and improved monitoring identity and metric attribution.
• 修复评估相关缺陷：任务列表多智能体过滤、AIDP 知识库用例生成支持、移除用例生成的损坏文件上传，并修复评估删除对租户管理员的回退逻辑。 / Fixed evaluation issues: task-list filtering for multiple agents, AIDP knowledge-base case generation, removal of the broken file-upload path, and tenant-admin fallback for evaluation deletion.
• 修复智能体配置更新特定场景下无限循环的问题。 / Fixed an infinite loop when agent configuration updates failed in certain scenarios.
• 修复 Agent 输出协议的重试与流式显示（develop 分支）。 / Fixed output-protocol retry and streaming display on the develop branch.
• 修复对话资源限制未本地化的问题，并拒绝超过 10 MB 的附件上传。 / Localized conversation resource limits and rejected attachments larger than 10 MB.
• 修复 AIDP 相关缺陷：折叠高级选项后权限丢失、文档列表历史合并、处理顺序与创建时间。 / Fixed AIDP issues: permissions lost when collapsing advanced options, and document-list history merge, processing order, and creation time.
• 修复知识库预览在轮询期间被刷新打断的问题。 / Fixed knowledge previews being refreshed during polling.
• 修复工作台创建智能体后的重定向与智能体配置页展示不全问题。 / Fixed redirection after creating an agent in the Workbench and incomplete rendering on the agent configuration page.
• 移除模型自动重命名行为，并加固配额、标签、模型与 API Key 等接口契约。 / Removed automatic model renaming and hardened quota, tag, model, and API-key contracts.
• 修复 OpenAI 兼容 provider 发现被 SSRF 防护误拦的问题，支持私网地址模型服务发现。 / Fixed the SSRF guard blocking OpenAI-compatible provider discovery, enabling privately addressed model services.
• 修复定时任务展示：改为显示下次执行时间并在任务结束后正确展示结果。 / Fixed scheduled-task display to show the next execution time and surface results after completion.

文档与部署变化 / Docs & Deployment Changes

• 合并 SQL 迁移为 v2.7.0_merged_migrations.sql，统一承载自 v2.6.1 以来的迁移内容（含智能体图标字段重命名等）。 / Consolidated SQL migrations into v2.7.0_merged_migrations.sql, carrying all migrations since v2.6.1 (including the agent icon column rename).
• 新增官方智能体部署脚本 deploy/deploy-official-agents.sh，支持一键导入预置智能体资产。 / Added deploy/deploy-official-agents.sh for one-click import of bundled official agent assets.
• 更新部署配置与环境变量模板：K8s Helm values/configmap、Docker Compose、构建脚本，新增租户资源限额配置项。 / Updated deployment configs and env templates (K8s Helm values/configmap, Docker Compose, build scripts) with new tenant resource-limit settings.
• 完善项目规范与测试资产：新增测试资产 skill 及配套参考文档。 / Expanded engineering standards and test assets with a new test-assets skill and its reference docs.

What's Changed

• ✨ Feature: support AIDP knowledge file deletion and download by @cj2026-bit in #3930
• Feature: Knowledge base interface optimization by @hzwllm in #3909
• 🐛 Fix: prevent preview refresh during file upload by @cj2026-bit in #3947
• perf(frontend): upgrade Next.js for faster Turbopack by @xuyaqist in #3949
• Bubfix: guarantee event order, harden chunk buffer, SSE-subscribe controller, and break adapter on terminal human_run by @bernard1234 in #3948
• Fix/override delete by @lijiayang619 in #3950
• 🐛 Fix(evaluation): fix task list filtering for multiple agents by @cj2026-bit in #3951
• fix(newchat): reject attachments larger than 10 MB by @xuyaqist in #3960
• 🐛 Bugfix: Fixed an issue where the sandbox container user lacked the permissions to create folders and files. by @YehongPan in #3964
• test(agents): fix base test by @xuyaqist in #3972
• 🐛 Bugfix: Fix issues with automated test cases. by @YehongPan in #3975
• [fix] 合并 CodeAgent 静默重试与显式终止协议 by @JasonW404 in #3973
• fix(hitl): harden HITL form delivery, event ordering and scheduler concurrency by @bernard1234 in #3977
• 🧪test: add test cases generaate skill & a2a mock services by @DongJiBao2001 in #3979
• fix(agent): accept reasoning-prefixed CodeAgent actions by @JasonW404 in #3989
• 0930需求-Nexent规格表：完善租户资源限制错误码、429响应及中英文提示 by @Junqilee in #3966
• refactor: unify resource page and resource card by @xuyaqist in #3992
• merge(develop): merge v2.6.1 hotfixes from hotfix/v2.6.1 by @jeffwu-1999 in #3998
• feat(agent-automation): show run result inline after cron job finished by @yunshuang-hw in #4000
• fix: harden quota, tag, model and API-key contracts found by daily test suite by @jeffwu-1999 in #4002
• fix(model): remove SSRF guard from OpenAI-compatible provider discovery by @lijiayang619 in #4004
• fix(agent-automation): show last execution time instead of next scheduled time by @yunshuang-hw in #4006
• ✨ Feature: Pass user context to MCP tools and A2A agents for tool-side authorization by @EDDIWARD in #3798
• bugfix:support model reasoning effort configuration by @hzwllm in #3953
• 🐛 Fix: upload flow bugs and evaluation delete tenant-admin fallback by @cj2026-bit in #3996
• bugfix: remove automatic model renaming by @hzwllm in #4009
• fix: normalize skill tags to prevent agents page crash by @MoeexT in #3991
• fix: quota breakdown pagination and platform quota panel hardening by @MoeexT in #3978
• fix: restrict model management to admins and hide model page from developers by @jeffwu-1999 in #4008
• bugfix: preserve AIDP permissions when submitting collapsed advanced options by @hzwllm in #4013
• ✨ Feature: Development of Agent Workbench Features by @YehongPan in #4010
• 🐛 Fix(evaluation): remove broken file upload from case generation, support AIDP knowledge bases by @cj2026-bit in #4012
• feat(logging):Security Audit Logging by @charmingchi1 in #4015
• 0930需求feat(agent): Nexent规格表-增加单租户智能体上限 by @Junqilee in #3701
• 0930需求feat: Nexent规格表-enforce conversation turn and history limits by @Junqilee in #3717
• 📦test: update a2a command document & update orjosn by @DongJiBao2001 in #4020
• Feat: bundled default agents by @menglinghan in #3899
• Xyq/refactor card by @xuyaqist in #4014
• feat(logging): 新增模型调用日志分类与路由能力 by @charmingchi1 in #3952
• Feat/model config redesign by @lijiayang619 in #4007
• 修复监控异常上报 by @hhhhsc701 in #3984
• 0930需求-feat: Nexent规格表- 增加租户知识库上限 + 知识库单文件大小上限 by @Junqilee in #3730
• 🐛 Bugfix: Added a configuration option to control the parallel_executor timeout. by @YehongPan in #4021
• feature: agent publish guide by @MoeexT in #4011
• bugfix: fix infinite loop when agent configuration update fails in certain scenarios by @hzwllm in #4022
• fix(agent): 修复输出协议重试与流式显示（develop） by @JasonW404 in #4023
• release(v2.7.0): sync main hotfix history, consolidate SQL migrations, bump version by @jeffwu-1999 in #4024
• Fix: AIDP document list history merge, processing order and creation time by @gs-aion in #4016
• feat: integrate official agent assets into Nexent repository by @menglinghan in #4027
• Optimize: superadmin management page by @xuyaqist in #4034
• 0930需求-Nexent规格表: （1）MCP：单租户MCP数量、超时时间；（2）Skill：单租户Skill数量、上传的Skill大小上限；（3）应用评测：单个评测集大小上限 by @Junqilee in #4025
• ♻️ Refactor: When launching the Agent Workbench, skip the home page and default directly to the Agent Workbench interface. by @YehongPan in #4036
• move Skill listing action to the More menu by @Summer-Si in #4038
• fix: show repository status in paged agent list by @xuyaqist in #4039
• 🐛 Bugfix: Fixed the issue with agent redirection when creating agents in the Agent Workbench; fixed the issue where agents were not fully displayed on the agent configuration page. by @YehongPan in #4040
• fix(frontend): handle agent name overflow and card tag layout by @xuyaqist in #4041
• fix: let tenant admins manage models of their own tenant by @jeffwu-1999 in #4043
• 🐛 Fix(evaluation): surface run delete errors and hide the delete button for unauthorized users by @cj2026-bit in #4042
• Fix/default model backfill select best by @lijiayang619 in #4045
• Release v2.7.0 merge by @jeffwu-1999 in #4047
• merge(main): merge v2.7.0 release from develop by @jeffwu-1999 in #4044

New Contributors

• @yunshuang-hw made their first contribution in #4000

Full Changelog: v2.6.1...v2.7.0