Skip to content

v0.8.0

Choose a tag to compare

@github-actions github-actions released this 07 Aug 06:10
· 17 commits to main since this release

禹都 AI 解决方案助手 v0.8.0

中文说明

本版本重点增强模型服务商兼容性、长任务稳定性、废标检查能力与本地使用体验。所有新增统计仍仅保存在本机,不接入远程埋点或远程 Analytics。

模型与服务商

  • 完善 Agnes AI 中国大陆站与国际站支持,内置 Agnes 2.0/2.5 Flash、2.5 Pro/Pro Alpha 及 Agnes Image 2.0/2.1 Flash。
  • 增强 DeepSeek 与 LongCat 模型配置,支持 Thinking 模式、推理强度和已知能力信息。
  • 新增模型能力探测、官方能力回退和本地缓存,模型列表仍由用户主动获取。
  • 优化生图模型质量、比例及兼容参数。

稳定性与安全

  • 增加统一的安全网络请求能力,包括超时、取消、响应大小限制和远程 URL 校验。
  • 加固 AI 生图、MinerU 上传下载、文档远程图片和 Word 导出图片链路。
  • 配置文件采用临时文件替换写入,降低异常退出造成配置损坏的风险。
  • 改进后台任务中断恢复、任务状态回放和全局任务监视器。
  • Mermaid 改为本地渲染与导出,提升稳定性并减少外部依赖。

投标与废标检查

  • 废标检查支持一次导入多份投标、资格、商务和技术文件。
  • 多文件合并分析时保留文件名,风险证据可以追溯到具体来源文件。
  • 新增“招标要求—投标响应—证据位置—状态—风险”符合性矩阵。
  • 优化响应文件要求提取、目录生成和后续正文生成的数据传递。

本地用量与界面

  • 新增本地 Token 用量统计,包括请求次数、输入、输出、Thinking 和总 Token。
  • 支持近 1 小时、6 小时、1 天、7 天和 14 天趋势曲线。
  • 新增模型用量排行、紧凑指标卡片和本地统计清理功能。
  • 调整设置页布局、Thinking Token 预算显示及提示样式。
  • 修复 Toast 生命周期与会话状态问题。

其他说明

  • 本版本不会新增远程统计、匿名标识或远程公告服务。
  • macOS 与 Windows 构建仍未接入代码签名,首次运行时可能出现系统安全提示。

English Release Notes

This release focuses on model-provider compatibility, long-running task reliability, bid compliance review, and local usability. All newly added usage statistics remain on the local device; no remote telemetry or analytics service is introduced.

Models and Providers

  • Improved Agnes AI support for Mainland China and international endpoints, with built-in presets for Agnes 2.0/2.5 Flash, 2.5 Pro/Pro Alpha, and Agnes Image 2.0/2.1 Flash.
  • Enhanced DeepSeek and LongCat configuration with Thinking mode, reasoning effort, and known capability metadata.
  • Added model capability probing, official capability fallback, and local caching. Remote model discovery remains user initiated.
  • Improved image-model quality, aspect-ratio, and compatibility parameters.

Reliability and Security

  • Added a shared secure HTTP layer with timeouts, cancellation, response-size limits, and remote URL validation.
  • Hardened AI image downloads, MinerU upload/download flows, remote document images, and Word export images.
  • Switched configuration persistence to temporary-file replacement to reduce corruption after unexpected shutdowns.
  • Improved interrupted-task recovery, task-state replay, and the global background-task monitor.
  • Moved Mermaid rendering and export to the local runtime for better reliability and fewer external dependencies.

Bidding and Compliance Review

  • Added multi-file import for bid, qualification, commercial, and technical documents.
  • Preserved source filenames during merged analysis so evidence can be traced to individual files.
  • Added a compliance matrix covering tender requirements, bid responses, evidence locations, status, and risk.
  • Improved response-file requirement extraction and propagation into outline and content generation.

Local Usage and UI

  • Added local Token usage statistics for requests, input, output, Thinking, and total Tokens.
  • Added trend ranges for the last 1 hour, 6 hours, 1 day, 7 days, and 14 days.
  • Added per-model rankings, compact metric cards, and local-statistics cleanup.
  • Improved settings layouts, Thinking Token budget presentation, and status messaging.
  • Fixed Toast lifecycle and session-state issues.

Notes

  • This release does not add remote telemetry, anonymous identifiers, or remote announcement services.
  • macOS and Windows builds are currently unsigned, so operating-system security warnings may appear on first launch.