Download Latest Version v2.5.0 source code.zip (58.3 MB)
Email in envelope

Get an email when there's a new version of Nexent

Home / v2.4.0
Name Modified Size InfoDownloads / Week
Parent folder
README.md 2026-08-05 25.0 kB
0
v2.4.0 source code.tar.gz 2026-08-05 54.3 MB
0
v2.4.0 source code.zip 2026-08-05 55.5 MB
0
Totals: 3 Items   109.7 MB 0

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

我们很高兴地宣布Nexent v2.4.0 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 v2.4.0 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

  • 新增智能体自动化任务，支持从对话识别意图、生成并编辑执行计划，以及管理定时调度和运行历史。 / Added agent automation tasks with conversational intent detection, editable execution proposals, scheduled runs, and run-history management.
  • 推出全新对话界面，增强附件、引用源、推理过程、工具调用和 Token 用量展示，并支持对话分享与 Mermaid 图表。 / Introduced a redesigned chat experience with richer attachment, citation, reasoning, tool-call, and token-usage views, plus conversation sharing and Mermaid diagrams.
  • 新增智能体规划与待办执行能力，支持并行调用多个子智能体并独立展示各路结果。 / Added agent planning and task-list execution, with parallel multi-sub-agent invocation and independently rendered results.
  • 引入统一的 ContextItem 上下文模型与自适应压缩机制，在长对话中更稳定地保留摘要、历史操作和关键资源。 / Introduced a unified ContextItem model and adaptive compaction to preserve summaries, action history, and key resources more reliably in long conversations.
  • 升级长期记忆架构，新增可扩展存储适配、记忆管理界面、定时整理，以及结合 MMR、时间衰减和 Token 预算的检索流程。 / Upgraded long-term memory with extensible storage adapters, a management UI, scheduled consolidation, and retrieval using MMR, temporal decay, and token budgeting.
  • 新增 Agent Guardrail 安全内容筛查和可控安全沙箱，为输入输出检查及工具执行提供更细粒度的防护。 / Added Agent Guardrail content screening and a controllable secure sandbox for finer-grained protection of inputs, outputs, and tool execution.
  • 新增 MCP 辅助的 NL2Agent 创建链路，可根据自然语言需求生成智能体草稿并推荐所需工具。 / Added an MCP-assisted NL2Agent workflow that turns natural-language requirements into agent drafts and recommended tools.
  • 扩展 Agent、MCP 与 Skill 资源库治理，支持 MCP 组权限和字段级分享、Skill 审核生命周期与快照，并将审核通知扩展到 MCP 和 Skill。 / Expanded Agent, MCP, and Skill repository governance with MCP group permissions and field-level sharing, Skill review lifecycles and snapshots, and review notifications for MCPs and Skills.
  • 增强知识库管理，新增容量配额与预警、可配置的上传限制，并完善 AIDP 知识库的创建、更新、文档管理和权限集成。 / Enhanced knowledge-base management with capacity quotas and warnings, configurable upload limits, and complete AIDP creation, update, document-management, and permission flows.
  • 扩展企业集成与定制能力，支持 A2A 自定义认证、W3 登录、自定义品牌信息以及 Web Base URL 构建。 / Expanded enterprise integration and customization with A2A custom authentication, W3 login, configurable branding, and Web Base URL builds.

Bug 修复 / Bug Fixes

  • 优化北向接口的会话与智能体查询，消除 N+1 查询并修正显式租户筛选，提升多租户场景下的性能和准确性。 / Optimized northbound session and agent queries by removing N+1 patterns and correcting explicit tenant filtering for better multi-tenant performance and accuracy.
  • 修复知识库的图片读取、文档预览、拖拽重复上传、嵌入模型匹配和容量达限后继续上传等问题。 / Fixed knowledge-base image retrieval, document previews, duplicate drag-and-drop uploads, embedding-model matching, and uploads after quota limits are reached.
  • 修复知识库、Skill 和智能体资源库中的权限与可见性问题，并防止认证 Token 泄露到任务结果。 / Fixed permission and visibility issues across knowledge bases, Skills, and the Agent repository, and prevented authentication tokens from leaking into task results.
  • 修复沙箱启用后 Agent 无法运行、ME VLM 调用失败、自验证配置未传递、并行执行序列化失败等运行时问题。 / Fixed runtime failures involving sandbox-enabled agents, ME-hosted VLM calls, missing self-verification configuration, and parallel-executor serialization.
  • 修复模型连通性校验与供应商保存的竞态、模型列表供应商错配、多模态嵌入模型无法添加等配置问题。 / Fixed race conditions in connectivity checks and provider saves, provider mismatches in model lists, and failures when adding multimodal embedding models.
  • 修复智能体配置页的 Prompt 同步、Skill 编辑重复 frontmatter、已删除 MCP 工具残留，以及智能体导入导出时的 Skill 丢失问题。 / Fixed prompt synchronization in agent configuration, duplicate Skill frontmatter, stale deleted MCP tools, and missing Skills during agent import and export.

文档与部署变化 / Docs & Deployment Changes

  • 重构中英文用户手册目录与图文内容，补充自动化任务、Agent/Skill/MCP 资源库、知识库、记忆、模型和新对话页指南，并支持文档图片点击放大。 / Reworked the bilingual user-guide structure and visuals for automation, Agent/Skill/MCP repositories, knowledge bases, memory, models, and the new chat experience, with click-to-zoom documentation images.
  • 更新 Docker、Kubernetes 与离线部署，统一容器 MCP 端口，新增沙箱镜像与配置，并调整 MinIO 镜像来源。 / Updated Docker, Kubernetes, and offline deployment by unifying container MCP ports, adding sandbox images and configuration, and changing the MinIO image source.
  • 按 v2.2、v2.3 和 v2.4 合并数据库迁移脚本，同步更新迁移说明与部署测试，简化新环境初始化和跨版本升级。 / Consolidated database migrations by v2.2, v2.3, and v2.4, with updated migration guidance and deployment tests for simpler fresh initialization and cross-version upgrades.

What's Changed

New Contributors

Full Changelog: https://github.com/ModelEngine-Group/nexent/compare/v2.3.0...v2.4.0

Source: README.md, updated 2026-08-05