Skip to content

LingShu v0.1.0-alpha.2

Pre-release
Pre-release

Choose a tag to compare

View all tags
@RoyZhao1991 RoyZhao1991 released this 17 Jul 20:12
v0.1.0-alpha.2
b929aed

LingShu v0.1.0-alpha.2

English | 简体中文

English

This is the first public alpha of LingShu, a native macOS AI agent that turns goals into verified deliverables. It is intended for developers and early adopters who are comfortable reviewing permissions, provider settings, and generated work before relying on it.

Highlights

  • Take a report brief end to end: create real editable presentation or document files, register and preview them locally, revise the actual output, and hand it to an independent checker instead of stopping at an outline.
  • Bring your own model through OpenAI, Anthropic Claude, DeepSeek, MiniMax M3, or a custom compatible endpoint.
  • Convert requests into structured goals, dispatch isolated workers, and verify results before completion.
  • Adapt the still-pending execution graph at runtime while keeping the original goal and acceptance boundary fixed.
  • Pause an exact worker or checker session for typed human collaboration such as login, QR scanning, file selection, confirmation, or physical action, then resume from that checkpoint.
  • Create, register, preview, and revisit real files such as documents, presentations, reports, code, and scripts.
  • Operate authorized Mac apps through native Accessibility-based Computer Use without requiring Codex.
  • Preserve task records, artifact history, and distilled child-task results across sessions.
  • Guide first-time users through brain-channel setup when no working model configuration is available.
  • Use the bundled lingshu CLI as a one-request/one-response entrance for Feishu bots, webhooks, Shortcuts, and local automation while preserving the same main-session memory, authorization, task records, and human-interaction checkpoints.

Install

  1. Download the Universal DMG and its .sha256 file from this release.
  2. Verify the download with shasum -a 256 -c LingShu-0.1.0-5-macOS-universal.dmg.sha256.
  3. Open the DMG and drag 灵枢.app to Applications.
  4. Launch LingShu and follow the brain setup guide.
  5. Grant macOS permissions only when a capability you choose requires them.

The optional CLI is bundled inside the installed app. See the CLI and connector guide for setup, JSON output, exit codes, and the external-connector pattern.

Homebrew users can install both the app and CLI with brew install --cask RoyZhao1991/tap/lingshu. Build 5 also fixes terminal-state reporting for one-request/one-response connectors, including compatibility with earlier cached task payloads.

The DMG supports Apple silicon and Intel Macs, is signed with a Developer ID certificate, notarized by Apple, and carries a stapled notarization ticket.

Package profile

To keep the first download deterministic and reasonably small, this DMG does not bundle the optional SenseVoice model runtime or the experimental HAL virtual microphone driver. Apple Speech remains available as the local speech-recognition fallback. The source tree retains both optional integration paths for development and later packages.

Requirements and known limitations

  • macOS 14 or later.
  • A token for a supported model provider, or a compatible custom endpoint.
  • Accessibility and Screen Recording are required only for the corresponding Computer Use capabilities; microphone, speech-recognition, and camera permissions are capability-specific.
  • LingShu is alpha software. Keep backups and review high-impact actions and generated deliverables.
  • Remote model and perception requests are governed by the configured provider's privacy and retention terms.
  • Provider behavior, multimodal support, permissions, and app accessibility trees vary by environment.
  • The HAL virtual microphone remains experimental and is not included in this DMG.

Please report reproducible defects with the bug template. Security issues should be reported privately according to the security policy.

简体中文

这是灵枢的首个公开 Alpha 版本。灵枢是一款把用户目标推进为可验收产物的原生 macOS AI Agent。本版本面向愿意检查系统权限、模型配置和生成结果的开发者与早期用户。

主要能力

  • 端到端完成汇报交付：把一条需求推进为真实可编辑的 PPT 或文档，在本地登记并预览，基于实际产物继续修改，再交给独立 checker 验收，而不是停在提纲。
  • 支持 OpenAI、Anthropic Claude、DeepSeek、MiniMax M3 及自定义兼容端点，主脑可替换。
  • 将请求整理为结构化目标，派发隔离工作线程，并在完成前进行结果验收。
  • 在不改变原始目标与验收边界的前提下，按执行现场动态调整尚未启动的工作流节点。
  • worker 或 checker 可因登录、扫码、选文件、确认或实体操作暂停精确会话，并在用户完成后从原断点继续。
  • 生成、登记、预览并持续追踪文档、PPT、报告、代码和脚本等真实文件。
  • 通过 macOS 原生辅助功能接口操作已授权应用，Computer Use 不依赖 Codex。
  • 跨会话保留任务记录、产物历史与子任务蒸馏结果。
  • 没有可用主脑配置时，首次启动引导用户完成通道设置。
  • 随 App 交付 lingshu CLI，为飞书机器人、Webhook、快捷指令与本地自动化提供一问一答入口，同时复用同一主会话的记忆、授权、任务记录和人机交互断点。

安装方式

  1. 下载本 Release 中的 Universal DMG 与对应 .sha256 文件。
  2. 运行 shasum -a 256 -c LingShu-0.1.0-5-macOS-universal.dmg.sha256 校验文件。
  3. 打开 DMG，将 灵枢.app 拖入“应用程序”。
  4. 启动灵枢，并按引导配置主脑通道。
  5. 仅在所选能力确实需要时授予对应 macOS 权限。

可选 CLI 已随安装后的 App 一并交付。安装方式、JSON 返回、退出码与外部连接器范式见 CLI 与连接器指南。

Homebrew 用户可通过 brew install --cask RoyZhao1991/tap/lingshu 同时安装 App 与 CLI。Build 5 修复了一问一答连接器的终态回报，并兼容早期版本缓存的任务载荷。

该 DMG 同时支持 Apple 芯片和 Intel Mac，使用 Developer ID 签名，已通过 Apple 公证并附加公证票据。

安装包说明

为保证首个下载包可复现且体积合理，本 DMG 不内置可选的 SenseVoice 模型运行时，也不携带仍处实验阶段的 HAL 虚拟麦克风驱动。Apple Speech 仍可作为本地语音识别回退方案；源码中继续保留这两条可选集成路径，供开发构建和后续安装包使用。

环境要求与已知限制

  • 需要 macOS 14 或更高版本。
  • 需要受支持模型服务商的 Token，或兼容的自定义模型端点。
  • 只有使用相应 Computer Use 能力时才需要辅助功能与屏幕录制权限；麦克风、语音识别和摄像头权限也按能力申请。
  • 灵枢仍处 Alpha 阶段。请保留重要文件备份，并检查高影响操作与生成产物。
  • 发送到远程模型或感知服务的内容受对应服务商隐私和留存条款约束。
  • 不同服务商的多模态能力、系统权限与应用辅助功能树可能存在差异。
  • HAL 虚拟麦克风仍是实验能力，本 DMG 不包含该驱动。

请使用 Bug 模板提交可复现问题；安全问题请按安全策略通过私密渠道报告。

Assets 5
Loading