VaneHub AI v1.2.0

VaneHub AI 1.2.0

选择语言 / Select language：

简体中文

VaneHub AI 1.2.0 是一个稳定桌面版本，为 Claude Code、Codex CLI、OpenCode、Gemini CLI、Antigravity CLI 以及内置 OnePiece API Agent 提供统一工作空间。

主要亮点

• 在一个工作空间中检测、安装、配置、启动并切换支持的编程 Agent。
• 统一管理会话、项目、Git worktree、本地终端和远程 SSH 工作空间。
• 管理 MCP Server、SDK、Skill、Prompt Hook、扩展、定时任务、通知和模型供应商配置。
• 查看用量、任务输出和经过脱敏的统一日志，同时遵循桌面服务边界。
• 原生桌面能力不可用时，可使用浏览器中的 Web/mock 运行时评估界面。

下载

平台	架构	安装包
Windows	x64	x64 .exe 安装程序（NSIS），已签署 updater；暂未提供 Authenticode 签名
macOS	Apple Silicon	aarch64 .dmg，已签署 updater；暂未进行代码签名和公证
macOS	Intel	x64 .dmg，已签署 updater；暂未进行代码签名和公证
Linux	x64	.deb 和 AppImage
Linux	ARM64	.deb 和 AppImage

本版本不提供 Windows ARM64、.msi 或 .rpm 安装包。Linux ARM64 安装包由 GitHub 的 ubuntu-24.04-arm 托管 runner 构建，该 runner 标签目前处于公开预览阶段。

验证下载

本阶段提供 Tauri updater 签名、安装包校验和、SBOM 和 GitHub 构建来源证明。Windows 安装包暂未使用 Authenticode 签名；macOS 安装包暂未使用 Developer ID 签名或公证；Linux 通过完整性和来源证明进行验证。

每个可下载安装包均包含在 SHA256SUMS 中，并提供 SPDX SBOM、GitHub build provenance 和 SBOM attestation。将 SHA256SUMS 与安装包下载到同一目录后运行：

sha256sum --check --ignore-missing SHA256SUMS

使用以下命令验证 GitHub 来源证明：

gh attestation verify <下载文件> --repo cdavid817/vanehub-ai

自动更新

稳定版使用已签名的 updater 产物和稳定更新通道。自动检查默认关闭，可在应用设置中启用。预览版使用独立通道，不会推送给稳定版用户。

已知限制

• 原生 CLI 检测、进程启动、本地存储、桌面集成和安装功能需要 Tauri 桌面应用；浏览器模式使用确定性的 Web/mock 行为。
• 因操作系统代码签名和公证将在后续阶段提供，Windows SmartScreen 和 macOS Gatekeeper 可能显示警告。
• Agent 厂商身份验证仍通过各厂商 CLI 或账户流程完成；VaneHub AI 不代理订阅登录。
• 可用安装包仅限于上表列出的平台和架构。

问题反馈

请通过 Bug 报告 提交版本号、操作系统和相关脱敏日志。产品建议请使用 功能请求。安全漏洞请通过 GitHub Security Advisory 私下报告，不要创建公开 Issue。

English

VaneHub AI 1.2.0 is the stable desktop release of the unified workspace for Claude Code, Codex CLI, OpenCode, Gemini CLI, Antigravity CLI, and the built-in OnePiece API agent.

Highlights

• Detect, install, configure, launch, and switch between supported coding agents from one workspace.
• Organize sessions, projects, Git worktrees, local terminals, and remote SSH workspaces.
• Manage MCP servers, SDKs, Skills, prompt hooks, extensions, scheduled tasks, notifications, and provider profiles.
• Inspect usage, task output, and redacted unified logs without bypassing the desktop service boundary.
• Use the browser-accessible Web/mock runtime for interface evaluation when native desktop capabilities are unavailable.

Downloads

Platform	Architecture	Assets
Windows	x64	x64 .exe installer (NSIS), updater-signed; Authenticode deferred
macOS	Apple Silicon	aarch64 .dmg, updater-signed; code signing/notarization deferred
macOS	Intel	x64 .dmg, updater-signed; code signing/notarization deferred
Linux	x64	.deb and AppImage
Linux	ARM64	.deb and AppImage

No Windows ARM64, .msi, or .rpm package is included in this release. Linux ARM64 packages are built on GitHub's ubuntu-24.04-arm hosted runner, whose label is currently in public preview.

Verify your download

This phase verifies the Tauri updater signature, package checksums, SBOM, and GitHub provenance. Windows packages are not Authenticode signed, and macOS packages are not Developer ID signed or notarized yet. Linux packages do not use operating-system code signing; their evidence is integrity and provenance only.

Every downloadable package is covered by SHA256SUMS, an SPDX SBOM, and GitHub build-provenance and SBOM attestations. Download SHA256SUMS beside the package and run:

sha256sum --check --ignore-missing SHA256SUMS

Verify GitHub provenance with:

gh attestation verify <downloaded-file> --repo cdavid817/vanehub-ai

Updates

Stable installations use signed updater artifacts and the stable update channel. Automatic checks are disabled by default and can be enabled from the application settings. Preview releases remain on a separate channel and are not offered to stable clients.

Known limitations

• Native CLI detection, process launch, local storage, desktop integration, and installation require the Tauri desktop application; browser mode uses deterministic Web/mock behavior.
• Windows SmartScreen and macOS Gatekeeper may warn because operating-system signing and notarization are deferred to a later release phase.
• Agent vendor authentication still occurs through each vendor's CLI or account flow; VaneHub AI does not broker subscription sign-in.
• Package availability is limited to the platform and architecture matrix above.

Reporting problems

Open a bug report with the release version, operating system, and a relevant redacted log excerpt. Use the feature request for product ideas. Report vulnerabilities privately through a GitHub security advisory, not a public issue.

---

What's Changed

Documentation

• docs: correct stable release version by @cdavid817 in #216
• Improve desktop usability, recovery, and startup experience by @cdavid817 in #220
• Multi-agent verification campaign: provider output resilience, live transcript rendering, and opt-in real-CLI desktop layers by @cdavid817 in #221
• feat(local-media): add offline OCR, STT, and TTS by @cdavid817 in #209
• Fix MCP runtime projection for desktop agent sessions by @cdavid817 in #222
• chore(release): prepare v1.2.0 by @cdavid817 in #225

Other changes

• test(skills): verify desktop agent effectiveness by @cdavid817 in #223
• Make the Loop definition wizard steer users to usable agents, with a desktop verification layer by @cdavid817 in #224

Full Changelog: v1.1.0...v1.2.0

VaneHub AI v1.1.0

VaneHub AI 1.1.0

VaneHub AI 1.1.0 is a stable desktop release of the unified workspace for Claude Code, Codex CLI, OpenCode, Gemini CLI, Antigravity CLI, and the built-in OnePiece API agent.

Highlights

• Detect, install, configure, launch, and switch between supported coding agents from one workspace.
• Organize sessions, projects, Git worktrees, local terminals, and remote SSH workspaces.
• Manage MCP servers, SDKs, Skills, prompt hooks, extensions, scheduled tasks, notifications, and provider profiles.
• Inspect usage, task output, and redacted unified logs without bypassing the desktop service boundary.
• Use the browser-accessible Web/mock runtime for interface evaluation when native desktop capabilities are unavailable.

Downloads

Platform	Architecture	Assets
Windows	x64	x64 .exe installer (NSIS), updater-signed; Authenticode deferred
macOS	Apple Silicon	aarch64 .dmg, updater-signed; code signing/notarization deferred
macOS	Intel	x64 .dmg, updater-signed; code signing/notarization deferred
Linux	x64	.deb and AppImage
Linux	ARM64	.deb and AppImage

No Windows ARM64, .msi, or .rpm package is included in this release. Linux ARM64 packages are built on GitHub's ubuntu-24.04-arm hosted runner, whose label is currently in public preview.

Verify your download

This phase verifies the Tauri updater signature, package checksums, SBOM, and GitHub provenance. Windows packages are not Authenticode signed, and macOS packages are not Developer ID signed or notarized yet. Linux packages do not use operating-system code signing; their evidence is integrity and provenance only.

Every downloadable package is covered by SHA256SUMS, an SPDX SBOM, and GitHub build-provenance and SBOM attestations. Download SHA256SUMS beside the package and run:

sha256sum --check --ignore-missing SHA256SUMS

Verify GitHub provenance with:

gh attestation verify <downloaded-file> --repo cdavid817/vanehub-ai

Updates

Stable installations use signed updater artifacts and the stable update channel. Automatic checks are disabled by default and can be enabled from the application settings. Preview releases remain on a separate channel and are not offered to stable clients.

Known limitations

• Native CLI detection, process launch, local storage, desktop integration, and installation require the Tauri desktop application; browser mode uses deterministic Web/mock behavior.
• Windows SmartScreen and macOS Gatekeeper may warn because operating-system signing and notarization are deferred to a later release phase.
• Agent vendor authentication still occurs through each vendor's CLI or account flow; VaneHub AI does not broker subscription sign-in.
• Package availability is limited to the platform and architecture matrix above.

Reporting problems

Open a bug report with the release version, operating system, and a relevant redacted log excerpt. Use the feature request for product ideas. Report vulnerabilities privately through a GitHub security advisory, not a public issue.

---

What's Changed

Documentation

• Improve agent settings UX and Skill synchronization by @cdavid817 in #196
• Make multi-Agent group chat work, and fix five more defects found on the desktop client by @cdavid817 in #197
• Refactor Web client state modules by @cdavid817 in #198
• Simplify the Prompt Hooks settings experience by @cdavid817 in #199
• Make human mention routing work in multi-Agent group chat, and isolate CLI config in desktop test runs by @cdavid817 in #201
• Cover the multi-Agent human decision pause and resume with a live desktop spec by @cdavid817 in #203
• Scope Claude permission hooks to VaneHub-managed sessions by @cdavid817 in #204
• Expand agent documentation and desktop verification by @cdavid817 in #200
• Refactor plan mode into OnePiece sessions by @cdavid817 in #202
• Archive managed permission hook scope change by @cdavid817 in #206
• feat: optimize loop engineering workbench by @cdavid817 in #205
• Make Agent evaluation ranking reach the client and failures explain themselves by @cdavid817 in #207
• Make the workspace scannable and recoverable by @cdavid817 in #208
• feat(cli-parameters): upgrade CLI parameter management by @cdavid817 in #210
• fix(test): stabilize desktop WebDriver worker lifecycle by @cdavid817 in #214
• chore(release): prepare v1.1.0 by @cdavid817 in #215

Dependencies

• deps(npm): bump expect-webdriverio from 5.6.5 to 6.0.5 by @dependabot[bot] in #213
• deps(npm): bump the npm-minor-patch group with 15 updates by @dependabot[bot] in #212

Other changes

• Include Linux ARM64 updater metadata by @cdavid817 in #195

Full Changelog: v1.0.0...v1.1.0

VaneHub AI v1.0.0

VaneHub AI 1.0.0

VaneHub AI 1.0.0 is the first stable desktop release of the unified workspace for Claude Code, Codex CLI, OpenCode, Gemini CLI, Antigravity CLI, and the built-in OnePiece API agent.

Highlights

• Detect, install, configure, launch, and switch between supported coding agents from one workspace.
• Organize sessions, projects, Git worktrees, local terminals, and remote SSH workspaces.
• Manage MCP servers, SDKs, Skills, prompt hooks, extensions, scheduled tasks, notifications, and provider profiles.
• Inspect usage, task output, and redacted unified logs without bypassing the desktop service boundary.
• Use the browser-accessible Web/mock runtime for interface evaluation when native desktop capabilities are unavailable.

Downloads

Platform	Architecture	Assets
Windows	x64	x64 .exe installer (NSIS), updater-signed; Authenticode deferred
macOS	Apple Silicon	aarch64 .dmg, updater-signed; code signing/notarization deferred
macOS	Intel	x64 .dmg, updater-signed; code signing/notarization deferred
Linux	x64	.deb and AppImage
Linux	ARM64	.deb and AppImage

No Windows ARM64, .msi, or .rpm package is included in this release. Linux ARM64 packages are built on GitHub's ubuntu-24.04-arm hosted runner, whose label is currently in public preview.

Verify your download

This phase verifies the Tauri updater signature, package checksums, SBOM, and GitHub provenance. Windows packages are not Authenticode signed, and macOS packages are not Developer ID signed or notarized yet. Linux packages do not use operating-system code signing; their evidence is integrity and provenance only.

Every downloadable package is covered by SHA256SUMS, an SPDX SBOM, and GitHub build-provenance and SBOM attestations. Download SHA256SUMS beside the package and run:

sha256sum --check --ignore-missing SHA256SUMS

Verify GitHub provenance with:

gh attestation verify <downloaded-file> --repo cdavid817/vanehub-ai

Updates

Stable installations use signed updater artifacts and the stable update channel. Automatic checks are disabled by default and can be enabled from the application settings. Preview releases remain on a separate channel and are not offered to stable clients.

Known limitations

• Native CLI detection, process launch, local storage, desktop integration, and installation require the Tauri desktop application; browser mode uses deterministic Web/mock behavior.
• Windows SmartScreen and macOS Gatekeeper may warn because operating-system signing and notarization are deferred to a later release phase.
• Agent vendor authentication still occurs through each vendor's CLI or account flow; VaneHub AI does not broker subscription sign-in.
• Package availability is limited to the platform and architecture matrix above.

Reporting problems

Open a bug report with the release version, operating system, and a relevant redacted log excerpt. Use the feature request for product ideas. Report vulnerabilities privately through a GitHub security advisory, not a public issue.

---

What's Changed

Documentation

• docs(zh): consolidate the Chinese documentation set and flatten README navigation by @cdavid817 in #102
• fix(release): publish a checksum manifest downloaders can verify by @cdavid817 in #105
• feat(plan): add Plan execution foundation by @cdavid817 in #106
• feat(agents): add Antigravity CLI as a fifth built-in CLI agent by @cdavid817 in #107
• chore(skills): refresh Claude workflow skills by @cdavid817 in #108
• feat: introduce extensible CLI provider contracts by @cdavid817 in #109
• feat: add workspace Tree-sitter code indexing by @cdavid817 in #110
• fix(skills): recover built-in Skill seeding from a registry/filesystem divergence by @cdavid817 in #111
• docs: surface project app icon in README hero by @cdavid817 in #112
• fix: harden runtime concurrency, transactional migrations, and query efficiency by @cdavid817 in #113
• feat: add durable session recovery by @cdavid817 in #117
• feat(lsp): add semantic code intelligence by @cdavid817 in #118
• feat: complete multi-agent session presence by @cdavid817 in #119
• feat(skills): add governed skill evolution foundations by @cdavid817 in #120
• Refine agent workflows, permissions, and tool activity by @cdavid817 in #121
• feat: complete OnePiece native Plan-Agent loop by @cdavid817 in #122
• feat: add fine-grained token accounting by @cdavid817 in #123
• Bind IM connectors to existing sessions by @cdavid817 in #124
• Add unified Todo Board by @cdavid817 in #125
• Add governed skill evolution evidence and utility delegation by @cdavid817 in #126
• feat: add OnePiece built-in tool system by @cdavid817 in #127
• Add OnePiece context management by @cdavid817 in #128
• Fix misleading workspace state and make destinations addressable by @cdavid817 in #129
• Add application-level slash commands for OnePiece sessions by @cdavid817 in #130
• Make file references work for source files by @cdavid817 in #133
• Add cross-platform desktop smoke automation by @cdavid817 in #132
• Archive the file reference preview and drag-paste changes by @cdavid817 in #134
• Expand the OnePiece native agent with five built-in capabilities by @cdavid817 in #136
• Add goals as a top-level entity over plans, loops, work items, and sessions by @cdavid817 in #138
• Archive the goal system change by @cdavid817 in #139
• Read and edit Jupyter notebooks a cell at a time by @cdavid817 in #140
• Stop direct connection from inheriting the operating system proxy by @cdavid817 in #137
• Rework cross-session memory into an addressable file store with two-tier recall by @cdavid817 in #141
• Archive the desktop smoke and slash command changes by @cdavid817 in #135
• Archive the direct-connection proxy inheritance fix by @cdavid817 in #142
• Advance Skill configuration, the sandboxed tool runtime, and Utility delegation by @cdavid817 in #143
• Enforce architecture fitness functions by @cdavid817 in #144
• Unify the Agent context engine by @cdavid817 in #145
• Add the agent code review center by @cdavid817 in #146
• Unify the agent run state machine by @cdavid817 in #147
• Add signed desktop releases and automatic updates by @cdavid817 in #148
• Add agent evaluation and benchmark platform by @cdavid817 in #155
• Add sandboxed Skill Tool runtime by @cdavid817 in #156
• Harden runtime lifecycle, persistence, and context boundaries by @cdavid817 in #157
• Rebuild the documentation topology and complete both user guides by @cdavid817 in #158
• Add Agent Mission Control by @cdavid817 in #159
• Extend the provider runtime plugin SDK by @cdavid817 in #161
• Make guide images resolve in the repository as well as the assembled site by @cdavid817 in #160
• perf: add runtime performance budgets by @cdavid817 in #162
• feat: add hybrid local model runtime by @cdavid817 in #163
• feat: add agent runner abstraction by @cdavid817 in #164
• Freeze line budgets for oversized source paths by @cdavid817 in #166
• Split migrations.rs into a directory module by @cdavid817 in #167
• Extract 102 methods from web-agent-client into bounded-context modules by @cdavid817 in #168
• Extract api_process_adapter inline tests to a child module by @cdavid817 in https://github.com/cdavid817/...

VaneHub AI v0.1.0-preview.1

This is a preview build. The packages below are unsigned and un-notarized. Your operating system will warn you before running them, and the steps to proceed are in Installing. Expect rough edges, and expect to reinstall rather than update — no automatic updater is configured.

What works

• Delivered: CLI management, single-Agent sessions, interactive Agent terminals, session organization, project/worktree and SSH workspace tools, settings, MCP/SDK/Skills/Prompt Hooks/extensions, IM connectors, scheduled tasks, notifications, usage reporting, and unified redacted logs.
• Preview: Multi-Agent coordination has native and Web/mock service contracts, but the create-session UI still disables Multi Agent mode.
• Not available yet: the Multi-Agent coordination UI, and Japanese application UI resources — Japanese currently covers the README only.
• Not available in this build: Claude Code permission-hook management. The hook is a separate binary that these packages do not yet ship, so enabling it in settings reports an error rather than writing a hook Claude Code could not run. Every other permission surface is unaffected, and your global Claude Code settings are left untouched.

Downloads

Platform	Asset	Notes
Windows x64	.exe installer (NSIS)	Installs per-user; no administrator rights required
macOS Apple Silicon	aarch64 .dmg
macOS Intel	x64 .dmg
Linux x64	.deb and AppImage	AppImage runs without installation

No .msi is published: the Windows Installer format cannot represent a pre-release version number. No .rpm is published: the RPM version field cannot contain the hyphen a pre-release version requires. Use the .exe installer and the AppImage respectively.

Installing

macOS

A .dmg downloaded through a browser is quarantined, and because this build is not notarized macOS reports that the application "is damaged and can't be opened". The application is not damaged — that is the message macOS uses for un-notarized software.

Drag the app to /Applications as usual, then clear the quarantine attribute:

xattr -cr "/Applications/VaneHub AI.app"

Open the application normally afterward.

Windows

SmartScreen shows "Windows protected your PC" because the installer carries no Authenticode signature. Choose More info, then Run anyway.

Linux

Install the .deb with your package manager, or mark the AppImage executable and run it directly:

chmod +x VaneHub-AI-*.AppImage
./VaneHub-AI-*.AppImage

Binaries are built on the current Ubuntu runner image and link against its glibc, so older distributions may refuse to start them.

Verifying your download

Every asset is listed in SHA256SUMS, published alongside them. Download it and check the file you retrieved:

sha256sum --check --ignore-missing SHA256SUMS

An SPDX SBOM and GitHub build-provenance attestations are also attached. You can verify provenance with:

gh attestation verify <downloaded-file> --repo cdavid817/vanehub-ai

Checksums, the SBOM, and attestations establish that these files came from this repository's build. They do not replace operating-system code signing or Apple notarization, and they do not remove the warnings described above.

Reporting problems

Open a bug report and include the version from this release, your operating system, and the relevant excerpt from the application logs. For a feature idea, use the feature request template. Please report security issues privately through a security advisory rather than a public issue.

---

What's Changed

Documentation

• chore: harden GitHub repository configuration by @cdavid817 in #1
• docs(readme): sync trilingual READMEs with delivered and planned features by @cdavid817 in #20
• feat: add Loop Engineering runtime and UI by @cdavid817 in #21
• feat(db): pool native SQLite connections by @cdavid817 in #17
• feat: add end-to-end execution observability by @cdavid817 in #22
• feat: optimize frontend rendering and lazy loading by @cdavid817 in #23
• test: establish coverage and reliability gates by @cdavid817 in #24
• feat: add multi-agent coordination runtime by @cdavid817 in #25
• build: optimize native linking and release packaging by @cdavid817 in #26
• feat: add advanced Prompt Hook library features by @cdavid817 in #27
• docs: establish multilingual documentation by @cdavid817 in #28
• feat: add remote terminal management by @cdavid817 in #29
• fix: render embedded terminals on an opaque dark canvas for TUI readability by @cdavid817 in #30
• fix: persist and resume exact CLI terminal sessions by @cdavid817 in #39
• feat: dynamic LLM model discovery from native CLI configs by @cdavid817 in #40
• fix: prevent settings hydration flash by @cdavid817 in #41
• fix: resolve session workspace UI defects found during tab audit by @cdavid817 in #45
• feat: add native API agent runtime with tools, compaction, skills by @cdavid817 in #46
• feat: track reported CLI token usage across managed and terminal sessions by @cdavid817 in #47
• feat: extend terminal usage tracking to gemini-cli by @cdavid817 in #48
• feat: add cross-session memory for native API agents by @cdavid817 in #49
• feat: add lifecycle management for native API agents by @cdavid817 in #50
• feat: add MCP tool-calling for native API agents by @cdavid817 in #51
• fix: preserve terminal usage integrity by @cdavid817 in #53
• feat: honor chat configuration for native API agents by @cdavid817 in #52
• feat: add tool-use trust setting for native API agents by @cdavid817 in #54
• feat: add Loop support for native API agents by @cdavid817 in #55
• chore: archive 9 completed custom API agent OpenSpec changes by @cdavid817 in #56
• chore: archive add-loop-api-agent-support OpenSpec change by @cdavid817 in #57
• feat: refine basic settings information architecture by @cdavid817 in #58
• perf: optimize runtime performance foundation by @cdavid817 in #59
• feat: add common application locales by @cdavid817 in #60
• feat: add global CLI agent configuration management by @cdavid817 in #61
• fix: harden skill and MCP management reliability by @cdavid817 in #62
• fix: repair Skill migration upgrade path by @cdavid817 in #63
• chore: archive completed OpenSpec changes by @cdavid817 in #64
• Fix/recover webview white screen by @cdavid817 in #65
• fix: harden MCP runtime reliability by @cdavid817 in #77
• feat: improve skill management and mount safety by @cdavid817 in #79
• feat: add native OnePiece agent and provider configuration by @cdavid817 in #83
• feat(im): harden connector runtime reliability by @cdavid817 in #86
• Polish workspace and settings visual consistency by @cdavid817 in #87
• feat(onepiece): add grep, glob, and edit tools and bound file reads by @cdavid817 in #88
• feat(personalization): custom instructions + cross-session memory by @cdavid817 in #89
• feat(retrieval): OnePiece 向量检索第 1 期 by @cdavid817 in #90
• docs: 让 AGENTS.md 的校验命令与 CI 实际 gate 一致 by @cdavid817 in #91
• chore: 建立 AI 编码助手的机器强制约束层 by @cdavid817 in #92
• feat(permissions): unified permission system across all managed CLI agents by @cdavid817 in #94
• docs: require English for commit messages and PR title/description by @cdavid817 in #95
• feat: multi-Agent group chat sessions with @ handoff by @cdavid817 in #96
• fix: harden native process, shell PTY, and session search paths by @cdavid817 in htt...