New version: Lime.Lime version 1.102.0 (#402052)

This commit is contained in:
spectopo authored and GitHub committed 2026-07-14 04:44:48 -07:00
1 parent 2fbdd5ac0b
commit 8a8a930b33
4 files changed
+127

No files matched your search

@@ -0,0 +1,24 @@
# Created with YamlCreate.ps1 Dumplings Mod
# yaml-language-server: $schema=https://aka.ms/winget-manifest.installer.1.12.0.schema.json
PackageIdentifier: Lime.Lime
PackageVersion: 1.102.0
InstallerType: exe
Scope: user
InstallModes:
- interactive
- silent
InstallerSwitches:
Silent: --silent
SilentWithProgress: --silent
UpgradeBehavior: install
Protocols:
- lime
ProductCode: lime
ReleaseDate: 2026-07-14
Installers:
- Architecture: x64
InstallerUrl: https://github.com/limecloud/lime/releases/download/v1.102.0/Lime-1.102.0.Setup.exe
InstallerSha256: F732614616820EC1A3F411F22BD9D17636FD15ADBBF0BC9152642B228E122609
ManifestType: installer
ManifestVersion: 1.12.0
@@ -0,0 +1,27 @@
# Created with YamlCreate.ps1 Dumplings Mod
# yaml-language-server: $schema=https://aka.ms/winget-manifest.locale.1.12.0.schema.json
PackageIdentifier: Lime.Lime
PackageVersion: 1.102.0
PackageLocale: en-US
License: Proprietary
ShortDescription: An open-source AI content workbench built for Chinese creators.
Description: |-
Lime is an open-source Electron desktop AI content workbench designed for Chinese creators, brand operators, research-oriented writers, and small teams. It covers AI writing, topic research, asset management, prompt accumulation, knowledge bases, and multi-model creative workflows.
You can think of it as an AI workspace better suited for long-term creation:
- It’s not just about single questions and answers; it’s about making continuous progress on a project.
- It’s not about finding references and rewriting prompts every time; it’s about accumulating references, styles, and methodologies.
- It’s not about results getting lost in chat history after generation; it’s about saving results so they can be reused next time.
- It’s not tied to a single AI service; it allows you to use the model capabilities you have already configured.
If you frequently switch between bookmarks, documents, chat tools, image tools, and model backends, Lime aims to consolidate these actions into a single creative space.
What you can do with Lime:
- AI Writing and Content Creation: Write WeChat Official Account articles, Xiaohongshu notes, video scripts, podcast outlines, and livestream scripts.
- Research and Knowledge Organization: Organize webpages, notes, screenshots, interviews, and historical materials to generate reports or briefs.
- Topic Analysis and Content Review: Deconstruct viral cases, competitor content, publishing cadences, and expression styles.
- Prompt Management and Style Reuse: Save common writing styles, brand tones, topic selection methods, and team templates.
- Multimodal Creative Preparation: Generate image prompts, cover directions, presentation structures, or webpage drafts.
- Multi-Model Workflow: Use your own configured AI providers and models to continuously edit, expand, compress, or cross-post within the same task.
ManifestType: locale
ManifestVersion: 1.12.0
@@ -0,0 +1,68 @@
# Created with YamlCreate.ps1 Dumplings Mod
# yaml-language-server: $schema=https://aka.ms/winget-manifest.defaultLocale.1.12.0.schema.json
PackageIdentifier: Lime.Lime
PackageVersion: 1.102.0
PackageLocale: zh-CN
Publisher: Lime
PublisherUrl: https://limeai.run/
PublisherSupportUrl: https://github.com/limecloud/lime/issues
PrivacyUrl: https://limeai.run/privacy/
PackageName: Lime
PackageUrl: https://limeai.run/
License: 专有软件
LicenseUrl: https://limeai.run/terms/
Copyright: © 2026 Lime
ShortDescription: 给中文创作者用的开源 AI 内容工作台
Description: |-
Lime 是一个开源的 Electron 桌面端 AI 内容工作台,面向中文创作者、品牌运营、研究型写作者和小团队,覆盖 AI 写作、选题研究、素材管理、提示词沉淀、知识库和多模型创作流程。
你可以把它理解成一个更适合长期创作的 AI 工作台:
- 不是只问一句、答一句,而是围绕一个项目持续推进
- 不是每次都重新找资料、重新写提示词,而是把参考、风格和做法沉淀下来
- 不是生成完就散落在聊天记录里,而是把结果保存起来,下一次还能继续用
- 不是绑定某一家 AI 服务,而是让你使用自己已经配置好的模型能力
如果你经常在收藏夹、文档、聊天工具、图片工具和模型后台之间来回切换,Lime 想帮你把这些动作收回到同一个创作空间里。
你可以用 Lime 做什么
- AI 写作与内容创作:写公众号文章、小红书笔记、视频脚本、播客提纲和直播话术
- 资料研究与知识整理:整理网页、笔记、截图、访谈和历史材料,生成报告或简报
- 选题分析与内容复盘:拆解爆款案例、竞品内容、发布节奏和表达风格
- 提示词管理与风格复用:保存常用写法、品牌语气、选题方法和团队模板
- 多模态创作准备:生成图片提示词、封面方向、演示稿结构或网页草稿
- 多模型工作流:使用自己配置的 AI 服务商和模型,在同一个任务里持续修改、扩写、压缩或换平台发布
ReleaseNotes: |-
Lime v1.102.0
新功能
- 打通 MCP server elicitation 的真实产品链:thread-scoped runtime 通过 App Server reverse JSON-RPC 发起表单请求,Electron 原样转发,Renderer 提供五语言交互,并支持 response、远端 resolved 与取消收口。
- 建立 durable Agent graph、identity 与 mailbox,并通过 per-turn gateway 向 current provider 提供 spawn_agent、list_agents、send_message、followup_task、interrupt_agent、wait_agent 六个控制工具;QueueOnly / TriggerTurn、canonical Item 和 delivered audit 共享同一持久化事实源。
- 补齐 Multi-Agent 按需恢复与终态通知:restart 后只 hydrate exact target,child 完成或失败通过 durable assistant Item / mailbox 通知 parent,wait_agent 以 steer-first 语义返回结构化 activity,并投影 canonical Collab / SubAgent Item。
- MCP Tool surface 改为 sampling-step snapshot,冻结 definition、allowlist、route、connection 与 timeout;资源和提示词读写使用精确 (server, uri/name) identity,避免跨连接误命中。
- 为 runtime MCP server 增加并发启动与故障隔离:optional server 失败不影响健康连接,required server 失败保持旧 generation 并拒绝不完整 replacement。
修复
- 修复 reverse JSON-RPC 的连接归属、boot-scoped request ID、abort cleanup、at-most-once response 与 resolved tombstone,重复、迟到、断连或未知响应继续 fail closed。
- 修复 MCP elicitation 的 active-time timeout、响应 _meta、表单 schema/number/enum 语义与取消竞态,用户等待时间不再计入服务端执行超时。
- 修复 Claw 首字与长工具循环稳定性:首轮响应策略统一归 App Server,移除 Renderer 快速路由双轨,并在已有联网证据后进入真实模型总结阶段。
- 修复 Claw 中断会话从“新建任务”首页重新进入同一会话时的导航去重,显式用户意图现在可以恢复 current read model、输入框与后续发送,不再被 stale draft pause 拦截。
- 修复媒体 provider/model 选择的原子持久化,以及侧边栏、归档会话、Automation、Skills 和 Plugin history 对旧 Agent runtime barrel 的依赖。
- 修复 approval session-cache、canonical Thread identity、active/queued Turn、终态历史与输入恢复之间的投影偏差,避免生成重复 pending 项或回退旧 read model。
优化与重构
- MCP management control plane 与 provider tool bridge 分离;canonical Tool Item 成为 history、compaction、coding、Skills、MCP、browser 与 artifact evidence 的唯一消费合同。
- 删除旧 Agent session store、subagent sidecar、aggregate runtime、session query、execution strategy compat、backend event stream 和默认 Playwright MCP seed,并补回流守卫。
- Agent Chat、Automation、Skills、归档会话、Sidebar 与 Plugin history 迁到各自 current owner,继续收缩旧兼容 barrel、快速响应 helper 和 pending shell 双轨。
- Multi-Agent canonical activity 收敛为 Started / Interacted / Interrupted 三类;list_agents 保持普通 Tool,旧 V1 resume / close 与历史 activity taxonomy 不再进入 current producer。
- 拆分 App Server event store、projection materializer、Agent runtime 与 GUI 大文件,保持领域 owner 和单一事实源边界。
测试与质量
- 扩展 App Server protocol/schema、Rust/TypeScript client、Electron host 与 Renderer contract,覆盖 server-originated request、resolved notification、MCP target identity 和 elicitation form。
- 新增 Agent graph/mailbox/control、restart-on-demand、terminal Result、concurrent wait、MCP router/timeout/fault isolation、canonical projection、first-token flow control 与 atomic media preference 的定向回归。
- 补充真实 Electron Gate B、Claw 中断后重入与继续会话、current fixture、协议守卫和 S1-S6 删除证据,验证生产 bridge 不回退 mock。
- 用户可见的 MCP elicitation 文案与 GUI 回归覆盖 zh-CN、zh-TW、en-US、ja-JP、ko-KR。
文档
- 更新架构、命令与 MCP 文档,明确 reverse JSON-RPC、MCP runtime/control-plane owner、Agent graph/mailbox、首轮响应策略和 canonical projection 边界。
- 更新 refactor v2 执行计划与逐切片 evidence,记录 current / compat / deprecated / dead 分类、验证结果和删除退出条件。
其他
- 版本事实源更新到 1.102.0:根应用、CLI npm package、Rust workspace、lime-rs/Cargo.lock 和 release notes。
完整变更:v1.101.0 → v1.102.0
ReleaseNotesUrl: https://github.com/limecloud/lime/releases/tag/v1.102.0
ManifestType: defaultLocale
ManifestVersion: 1.12.0
@@ -0,0 +1,8 @@
# Created with YamlCreate.ps1 Dumplings Mod
# yaml-language-server: $schema=https://aka.ms/winget-manifest.version.1.12.0.schema.json
PackageIdentifier: Lime.Lime
PackageVersion: 1.102.0
DefaultLocale: zh-CN
ManifestType: version
ManifestVersion: 1.12.0