The assistant workbar now uses persistent application tabs with a permanent catalog entry point and supports multiple local or managed SSH terminals bound to the current project. Terminal sessions flow through the Main, preload, and renderer boundaries with bounded output, resize, lifecycle, and cleanup behavior.
The main sidebar and Magic Notes list now support pointer and keyboard resizing. Magic Notes can also hide and restore its list, remembers the chosen layout, and stacks its panes when the available width is narrow.
Closing a terminal tab ends its Shell session. Restored terminal tabs do not restart prior Shell processes and require an explicit reconnect.
Release note: 助手工作栏现可通过“+”打开多个本机或当前远程项目终端；主侧栏和魔法笔记列表支持拖拽调宽，魔法笔记列表可隐藏并记住布局。
Managed SSH prompts previously depended on Desktop-owned provider and transcript state, so accepted work could stop or become uncertain when the Desktop, network, or power disappeared. The Agent now owns accepted OpenCode prompt execution, model and tool rounds, cancellation, and the durable semantic transcript, while Desktop remains authoritative for permanent projects, conversations, tasks, and messages.
Recovered work attaches to the original conversation without replay. The project switcher also groups managed SSH projects by saved Host and shows live Agent connectivity beside each group without opening a connection merely to read status. Remote project subtitles now show only the path, and the Agent lock advances to 0.11.14 with Runtime ACP v4.
Release note: 托管 SSH 任务在桌面端、网络或电源中断后可继续在主机上运行，并能回到原会话精确恢复；项目菜单现按主机分组并显示实时连接状态。
OpenCode Task calls now appear as first-class subagents in local and managed SSH conversations, with stable status, persistence, terminal handling, and true event ordering instead of duplicate generic tool entries.
GoodBuddy no longer applies a fixed outer wall-clock total to OpenCode, Continue, DeepSeek Harness, or managed SSH Prompts. Runtime-owned limits, user cancellation, bounded control operations, output limits, and released Agent protocol compatibility remain intact. Desktop and Agent release metadata are synchronized at 0.11.13 with approved bilingual notes.
Release note: OpenCode 原生 Task 现在以独立子 Agent 展示；长任务不再被 GoodBuddy 固定总时限中止，并修复 OpenCode 启动识别、可配置 HTTP 服务地址与发布说明读取问题。
DeepSeek Harness and Weixin binding configuration rejected non-loopback HTTP endpoints even when users explicitly selected those internal services. Configurable endpoints now accept valid HTTP and HTTPS URLs without host-location restrictions, while retaining the required protocol and API-key checks.
Release note: 模型连接、DeepSeek Harness 与微信绑定现在可直接使用局域网或公网 HTTP/HTTPS 服务地址，不再强制公网 HTTPS 或本机回环地址。
Managed SSH project selection repeatedly activated the remote Host and could
leave idle ACP bindings rejected as stale-controller after a controller
generation change. Host environment updates also coupled project state to
component installation evidence and presented a crowded control surface.
Project selection is now local-only, one Host connection serves compatible
project and conversation callers, idle bindings are safely replaced only
after proven controller takeover, and active prompts keep exact resume/replay.
Remote environment setup now uses explicit signed compound package
preparation with bounded transfer, rollback, recovery, and clearer settings.
Managed Hosts must be updated to separately distributed Agent 0.11.6 before
the daemon-side controller recovery fix takes effect. Desktop packages do not
embed the Agent or remote Runtime, and existing project data is preserved.
Release note: 托管 SSH 项目切换不再重复连接，同一 Host 可顺畅承载多个项目和对话；同时修复重连后的 stale-controller，并优化远程环境安装、更新与状态布局。
Direct Anthropic connections reused the configured context window as max_tokens, which could impose an unintended output value or send a provider-invalid request. Anthropic connections now expose an independent per-response maximum output setting, defaulting to 32K for the protocol-required field, while OpenAI direct requests remain without an added output token cap and direct tool rounds remain unmetered.
Release note: Anthropic 直连模型现在可单独配置单次最大输出，不再复用上下文窗口；OpenAI 直连请求和工具调用轮次不增加额外配额。
Valid Agent packages could be rejected when packaged trusted-key JSON used CRLF, and Settings could not show whether a newer compatible Agent existed. Registry formatting is now parsed independently of cryptographic trust, and Settings reads only the signed catalog to report local and latest versions.
New installations use a credential-free local Ollama-compatible profile. The v20 migration changes only untouched historical defaults, while explicit profiles and legacy environment aliases remain compatible. Magic Notes count badges no longer shift navigation rows.
Release note: 修复 Agent 下载兼容性并在设置中显示更新状态；新安装默认连接本地模型，保留显式用户配置；魔法笔记待办数字不再改变侧栏菜单高度。
Clicking the built-in embedding model download could appear to do nothing while
Main waited for the full transfer to finish. The vector model UI now follows
the speech model workflow with an immediate preparing state, verified source
and package metadata, live progress, cancellation, unavailable-source guidance,
and success notifications.
Use a bounded progress-only IPC instead of overlapping full snapshot polling,
and share model package size and progress presentation across embedding,
speech, and OCR model settings. Promote the GoodBuddy Agent lock from the e2e
prerelease identifier to the formal 0.11.2 version for production publication.
Release note: 修复点击内置向量模型下载后界面无响应的问题；现在会立即显示下载来源、进度与取消操作，并将 GoodBuddy Agent 版本标识转为正式的 0.11.2。
Managed SSH previously tied remote Agent and Runtime artifacts to desktop
packaging and Host activation. Settings now manages verified Linux x64 and
arm64 .gbagent packages explicitly, with source-aligned download and update
plus offline import and export. Each package contains the signed Agent, pinned
Node, and compatible OpenCode Runtime, and project activation never downloads
one automatically.
Dedicated native CI and a protected release workflow publish deterministic,
signed cumulative catalogs to GitHub and Beijing OSS without replacing the
desktop Latest release. Desktop packaging is metadata-only, remains independent
of Agent artifacts and signing credentials, and Linux desktop releases add RPM
installers.
The managed SSH path also keeps model credentials in Main, enforces Ask as
read-only, verifies signed Agent and Runtime inputs, preserves ACP recovery, and
reports Host update cancellation and failures accurately.
Release note: 远程项目现可在“设置”中手动下载、更新或离线导入签名 Agent 包，桌面安装包与远程组件发布相互独立；Linux 桌面版同时新增 RPM 安装包。
GoodBuddy previously had no direct way to send feedback from the desktop app. The About & Updates page now opens a bilingual, accessible feedback dialog for problems, feature requests, experience notes, and other feedback, with an optional email address and one screenshot.
Submissions use the narrow preload and trusted IPC boundary, a fixed production endpoint with strict TLS, bounded screenshot processing, an isolated resettable installation ID, preserved drafts, and idempotent manual retries. GoodBuddy does not automatically attach conversations, logs, files, settings, or credentials.
Release note: 现在可在“关于与更新”中直接提交问题、建议与体验反馈；邮箱和截图均可选，提交失败时会保留草稿供重试。
Add managed SSH Hosts and projects with signed side-by-side GoodBuddy Agent and OpenCode Runtime installation, Ask and Execute workspace access, a Main-owned model bridge, detached Agent lifecycle, durable ACP output recovery, and project activation UI.
Use protocol v2 with bounded FIFO frames over SSH instead of custom credit, priority, batching, or model-message chunking. Mismatched Host Agents now upgrade automatically, an exactly verified Node runtime is reused on-host, and local Portable builds can embed strictly verified Agent artifacts.
Release note: 新增托管 SSH 远程项目，可自动安装或升级签名 Agent/OpenCode Runtime，并在 SSH 断线后继续 Agent 工作和同步未确认输出。
GoodBuddy previously required an external OpenAI-compatible endpoint for vector retrieval. It can now install, verify, import, remove, and run the built-in Granite multilingual embedding model in an isolated local inference process, while retaining custom embedding connections and migrating existing settings.
The model connection UI now manages built-in and custom vector connections consistently, reports the actual retrieval path, preserves credentials across failed saves, and renders the vector tab synchronously. Related settings copy, switch styling, metadata typography, and renderer bundle guardrails are aligned with the resulting workflow.
Release note: 新增可管理的内置 Granite 多语言向量模型，支持本地下载、离线导入、连接切换与诊断；同时优化向量设置及相关界面的文案、开关和状态显示。
Magic Notes can now show a default-on incomplete todo badge in the primary
navigation. The badge hides at zero, caps its visual label at 99+ while
retaining the exact accessible count, and refreshes after user or Agent/MCP
note changes.
Platform Features > Magic Notes includes a persistent switch that hides only
the count without changing notes or the Magic Notes entry. Existing settings
migrate with the count enabled.
Release note: 左侧“魔法笔记”入口现在可显示未完成待办数量；可在“平台功能 → 魔法笔记”中关闭，超过 99 项显示“99+”。
Completing a Magic Notes todo updated the stored checklist but left the cached
source note stale until the workspace was remounted. Todo updates now return
the refreshed source note and update the active note cache immediately for
both completion and reopening.
Release note: 修复在待办视图勾选或取消待办后，源笔记必须切换页面才显示最新状态的问题；现在会即时同步。
Local conversations can now continue in a new branch after stable content is persisted. Each branch uses new conversation and message IDs, keeps its source visible, and remains independent without copying Tasks, queue items, runs, memories, or artifact ownership.
Conversation Task rows are also denser, align status markers at the right edge, and use an accessible checked success marker without redundant completion text.
Release note: 本地会话现在可从“在新会话中继续”创建独立分支并显示来源；Task 子项布局更紧凑，状态统一显示在右侧。
Address reliability and consistency gaps across Agent Runtimes, persistence, settings, Knowledge, Magic Notes, Smart Heartbeat, and the download site. Runtime processes now have bounded lifecycle cleanup and atomic configuration rollback, while model packages and persisted mutations recover safely.
Add a configurable global shortcut, protect unsaved work, improve modal and keyboard behavior, localize the built-in project without rewriting stored data, and lazy-load heavy renderer routes under enforced bundle budgets. Align project forms and disabled controls with shared typography and interaction states, and strengthen website release metadata validation and navigation accessibility.
Release note: 修复 Runtime、设置、知识库、魔法笔记与智能心跳中的可靠性和交互一致性问题；新增可配置全局快捷键，改进无障碍与加载性能，并强化官网下载校验。
Messages and Scheduled Task occurrences previously could not share one ordered path while a Conversation was active. They now enter a durable FIFO queue with frozen execution settings and bounded attachments, recover safely after restart, and never write concurrently to the same timeline.
The compact queue above Composer supports removal and explicit interrupt-and-promote actions. SQLite schema v23 preserves pending schedule work, while Conversation Task children reuse the shared status-dot semantics for running, completed, failed, approval, paused, and cancelled states.
Release note: 回复生成期间仍可继续发送普通消息；消息与 Scheduled Task 会按 Conversation 顺序排队，并支持删除待发送项或立即中断后优先执行。
Managed speech and OCR downloads previously used catalog-specific source URLs without a global selection. Platform Features now lets users choose ModelScope by default or Hugging Face, while Main validates and freezes that source for each download.
Verified coverage remains explicit: downloads never mix artifacts or silently switch sources, and installed models plus ZIP imports stay source-independent.
Release note: 可在“设置 → 平台功能 → 通用设置”中选择 ModelScope 或 Hugging Face 作为后续语音输入与 OCR 模型下载源；缺少完整已验证文件或下载失败时不会静默换源。
Parallel expert runs previously exposed only child status, while the synthesized answer appeared above the expert cards. Each expert now retains and exposes its complete output, the cards can be expanded, and the final synthesis is rendered beneath them and persisted with the conversation.
Release note: 多专家并行分析现在可展开查看每位专家的完整输出，并在其下方显示总 Agent 的综合结果。
Built-in MCP servers were always granted to supported runtimes without per-server controls. They now have persistent enablement and runtime assignments for direct models, managed OpenCode, and Continue, while DeepSeek Harness remains visibly unsupported and cannot be assigned.
MCP settings are reorganized into Built-in MCP, Direct model, Custom MCP, and Computer control tabs. Direct-model web search now uses the same accessible collapsible inventory pattern as other tool groups.
Release note: 内置 MCP 现在可分别启停并分配给直连模型、OpenCode 和 Continue；MCP 设置分类与直连模型工具列表也更清晰，DeepSeek Harness 会明确显示为暂不支持。
OpenCode and Continue customization was previously planned but unavailable, and Runtime-native capabilities were not represented consistently across providers. GoodBuddy now provides secure Main-owned customization settings, truthful native inventories, OpenCode Agents and Commands, Continue Rules and Prompts, DSH Web Search/Fetch, MCP Prompt and Resource metadata, and manual compaction where supported.
Native capabilities are presented in eleven accessible tabs with Tools separated from Commands, LSP, and Formatters. Tool source and Ask/Execute availability are explicit, external OpenCode remains connection-only, Continue reports unsupported static tool discovery instead of advertising unreachable Skills, and disposable inventory probes avoid retaining background runtimes.
Ask remains read-only at the Runtime boundary, Execute keeps the existing authorization controls, and credentials remain confined to Main.
Release note: 新增 OpenCode、Continue 与 DeepSeek Harness 的 Runtime 原生定制与真实能力清单；工具来源、Ask/Execute 可用性、上下文压缩和 MCP 元数据现在可清晰查看，同时继续保持 Main 进程凭据保护与现有权限边界。
GoodBuddy could share Skills across runtimes, but custom MCP remained limited and DeepSeek Harness could not manage third-party extensions. The app now provides a default-off DSH npm marketplace with managed installation, configuration, failure isolation, and packaged npm support, while assigned custom MCP is available to managed OpenCode, Continue Agent, and DeepSeek Harness in Execute.
Third-party DSH install scripts, initialization, and tools run with the current user's permissions. Ask remains read-only at dispatch, and turning off the marketplace hides management without disabling installed plugins.
Release note: 新增默认关闭的 DSH 插件市场，并让自定义 MCP 可分配给 OpenCode、Continue 和 DeepSeek Harness；安装第三方插件前会明确提示当前用户权限边界。
Long direct-model conversations mixed provider usage with local estimates,
and Agent tool rounds could remain above the configured compression target.
Compression state and status markers also did not reliably survive restarts
or bounded history rollover.
Direct-model calls now prefer provider-reported usage, compact complete
conversation turns and Agent tool rounds within reserved payload budgets,
and persist reusable summaries with scope-specific markers. The chat meter
separates latest-call usage from estimated compressed conversation size,
while failed or cancelled calls retain the last successful measurement.
Release note: 直连模型现可在长对话和多轮工具执行中自动压缩旧上下文，并分别显示本次调用用量与压缩后对话估算；摘要会自动保存并跨重启复用，无需手动操作。
mesalogo