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 / v2.5.0
Name Modified Size InfoDownloads / Week
Parent folder
README.md 2026-08-29 29.7 kB
0
v2.5.0 source code.tar.gz 2026-08-29 56.9 MB
1 1 weekly downloads
v2.5.0 source code.zip 2026-08-29 58.3 MB
0
Totals: 3 Items   115.2 MB 1

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

我们很高兴地宣布Nexent v2.5.0 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 v2.5.0 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

  • 重构 Agent 配置与 NL2Agent 生成流程，支持从自然语言需求创建 Agent、发现并安装可用资源，以及推荐和预填已绑定工具。 / Refactored Agent configuration and NL2Agent generation to create agents from natural-language requirements, discover and install available resources, and recommend and prefill bound tools.
  • 新增完整的 Agent 评估工作台，覆盖评估集、评估器、标注、调度执行与中英文报告，并提供通用基准测试和上下文对比工具。 / Added an end-to-end agent evaluation workspace covering datasets, evaluators, annotations, scheduled execution, and bilingual reports, together with generic benchmarking and context-comparison tools.
  • 新增长期记忆 Dreaming 与版本化管理能力，并提供记忆配置、检索记录和可视化管理界面。 / Added long-term memory Dreaming and version management, with configuration, retrieval records, and visual management interfaces.
  • 扩展运行时沙箱，支持会话级与系统级隔离、默认启用系统级沙箱、在沙箱内执行 Skill 脚本，并在前端展示生成文件。 / Expanded runtime sandboxing with session-level and system-level isolation, system sandboxing enabled by default, Skill script execution inside sandboxes, and generated-file display in the frontend.
  • 新增 S3 文件上传下载工具并增强文件处理链路，使 Skill 和工具能够直接处理本地文件与产物。 / Added S3 upload and download tools and enhanced the file-processing pipeline so Skills and tools can work directly with local files and generated artifacts.
  • 扩展统一模型网关，接入 LLM、Embedding、Rerank 及图像、视频、音频理解能力，并增强 ASR、结构化 VLM 消息、Logprobs 与瞬态错误重试支持。 / Expanded the unified model gateway across LLM, embedding, reranking, and image, video, and audio understanding, with improved ASR, structured VLM messages, logprobs, and transient-error retries.
  • 增强知识库与资源治理，支持个人知识库容量、租户资源限额、会话级知识范围选择及更灵活的文件切分与处理。 / Enhanced knowledge-base and resource governance with personal capacity controls, tenant quotas, conversation-level knowledge scoping, and more flexible file splitting and processing.
  • 优化对话体验，支持消息时间戳、按需分页加载、批量删除会话，并增强新旧聊天界面的运行时协同。 / Improved conversations with message timestamps, on-demand pagination, bulk deletion, and stronger runtime coordination across the new and legacy chat interfaces.
  • 扩展平台开放能力，新增 API Key 管理、北向 /models 与 /generate_title 接口，并补充运行接口的会话创建事件。 / Expanded platform APIs with API key management, northbound /models and /generate_title endpoints, and conversation-creation events in the run API.
  • 新增上下文与缓存监控指标，并强化元数据传递和提示注入防护，提升 Agent 运行的可观测性与安全性。 / Added context and cache metrics while strengthening metadata propagation and prompt-injection protection for more observable and secure agent execution.

Bug 修复 / Bug Fixes

  • 修复知识库导入死锁、不可重试错误未及时终止、文件删除与预览异常、EPUB 处理失败及数值检索排序问题。 / Fixed knowledge-base import deadlocks, non-retryable failures that did not terminate promptly, file deletion and preview errors, EPUB processing failures, and numeric retrieval ranking.
  • 修复 NL2Agent 表单锁定与工具配置问题，并增强 Skill 编码、中文名称、重命名元数据及 Agent 导入复制时的冲突处理。 / Fixed NL2Agent form locking and tool configuration, and improved Skill encoding, Chinese names, rename metadata, and conflict handling during agent import and copy.
  • 修复会话删除、重命名、恢复及分页后空列表问题，并解决流式响应缓存、空最终回答和运行时内存压力异常。 / Fixed conversation deletion, renaming, resume, and empty lists after pagination, along with streaming cache issues, empty final answers, and runtime memory pressure.
  • 修复评估任务在沙箱中的执行、中文评估报告、场景描述校验及评估路由和权限配置问题。 / Fixed evaluation execution in sandboxes, Chinese evaluation reports, scenario-description validation, and evaluation routing and permission configuration.
  • 修复 MCP 状态显示、工具刷新、Registry 删除及配置页同步问题，同时完善容器启动日志展示。 / Fixed MCP status display, tool refresh, Registry deletion, and configuration-page synchronization, while improving container startup log visibility.
  • 修复定时任务无法创建、模型管理会话过期处理以及外部 A2A 子 Agent 凭据传递等运行与鉴权问题。 / Fixed scheduled-task creation, expired-session handling in model management, and credential propagation to external A2A sub-agents.

文档与部署变化 / Docs & Deployment Changes

  • 更新 Docker、Helm、离线包与镜像构建配置，改进 MinIO 数据持久化和就绪检查，并保持环境变量配置兼容性。 / Updated Docker, Helm, offline-package, and image-build configurations, improved MinIO persistence and readiness checks, and preserved environment-variable compatibility.
  • 汇总 v2.5.0 SQL 迁移并增强合并迁移的重复应用逻辑，同时补充不可变迁移文件规范。 / Consolidated v2.5.0 SQL migrations, improved reapplication of changed merged migrations, and documented the immutable-migration policy.
  • 更新安装、Kubernetes、主页、资源管理和官方文件生成 Skill 文档，并补充账户初始化与自定义文件生成指南。 / Updated installation, Kubernetes, home-page, resource-management, and official file-generation Skill documentation, with new account-setup and custom file-generation guides.

What's Changed

New Contributors

Full Changelog: https://github.com/ModelEngine-Group/nexent/compare/v2.4.1...v2.5.0

Source: README.md, updated 2026-08-29