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。
Legacy key-only environment deployments retain their historical model endpoint, name, and protocol while generic model variables continue to take precedence. Agent inventory scanning now overlaps the signed catalog check instead of waiting for the network first.
The separate LoongArch preview builder derives node-pty from the candidate lockfile and verifies the installed version before cross-compiling it, preventing stale preview-only pins.
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。
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。
Save and test could accept a successful endpoint response without proving that the selected model generated output, while changing a connection URL or temporarily disabling authentication could require the API Key again.
Model tests now issue bounded real text or image generation requests and validate their output. API Keys remain encrypted with their connection across URL and authentication changes until explicitly cleared or the connection is deleted, and the form places the key directly below authentication. Real tests may incur a small amount of provider usage.
Release note: 修复模型“保存并测试”仅验证连通性的问题；现在会执行真实文本或图片生成测试，并在修改地址或临时关闭认证时继续保留该连接的加密 API Key。
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。
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。
Make agent backends configurable and governable with encrypted credentials, explicit context sharing, and tool approval boundaries.
Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
mesalogo