lbjlaq/Antigravity-Manager v5.0.0-beta.0

Antigravity Tools v5.0.0-beta.0

on GitHub

• [Architectural Decoupling from Upstream quotaInfo Omission & Full-Spectrum Gemini Resilience] (Fixes #3737, Thanks to @xiaozhi-xmu, @chenshin166-arch):
  • Decoupled Model Ingestion from quotaInfo: Completely eliminated the legacy if let Some(quota_info) = info.quota_info filter gate in quota.rs, benchmarking exclusively against official model catalogs and baseline compliance to retain Gemini models (gemini-3.8-flash-*, gemini-3.1-flash-lite, etc.) when upstream omits model-level quotaInfo, permanently resolving local 503 scheduling rejections. (Fixes #3737, Thanks to @xiaozhi-xmu, @chenshin166-arch)
  • Pre-Fetch Group Quota & Short-Board Fused Binding: Shifted retrieveUserQuotaSummary retrieval ahead of model parsing to seamlessly bind 5h and weekly quota for quotaInfo-less models; Argon prioritizes standalone groups and gracefully falls back to Gemini groups. (Fixes #3737)
  • Zero-Quota Guard & Latest Reset Time Protection: Safeguards explicit 0% model-level quota against positive group overrides; binds the later reset time when 5h and weekly quotas are identical to protect account dispatchability. (Fixes #3737)
  • Deprecate Obsolete gemini-pro-agent Codename: Standardized gemini-3.1-pro Medium and High tiers in variant_mapping.rs to official physical model gemini-3.1-pro-high, eradicating 503 No accounts available with quota for model: gemini-pro-agent errors. (Fixes #3737)
  • Claude Parity for Dynamic Model Candidate Rewriting: Integrated token_manager.resolve_dynamic_model_for_account into the Claude handler post-scheduling to match OpenAI and Gemini parity for adaptive per-account model rewriting. (Fixes #3737)
• [Gemini 4 Argon Dispatch Closure & Standalone Quota Anti-Bleed Isolation] (Fixes #3735):
  • Scheduler Convergence Alignment for Advanced Models: Extended candidate filtering in TokenManager::get_token_internal and fallback routines with token_supports_advanced_model and standalone bucket checks (normalized_target != "claude" && token.model_quotas.contains_key(...)), eliminating the fatal phase-two rejection of Argon-capable accounts while strictly preserving Claude 5.5 and 4.6 pool separation. (Fixes #3735)
  • Robust Quota Group Ingestion & In-Memory Hydration: Supported both camelCase and snake_case properties across load_single_account, get_available_models_from_json, and quota.rs, reliably capturing Argon quota whenever group or bucket names match argon; automatically hydrates in-memory available_models and limit entries whenever an account holds gemini-4-argon. (Fixes #3735)
  • Streamline Account Table Email Layout & Prevent Badge Overflow: Refactored the email cell in AccountTable.tsx to a vertical flex container and enabled badge chips to wrap cleanly with flex-wrap and removed shrink-0, preventing badge clusters from occluding adjacent quota progress cards. (Fixes #3735)
• [Mini Floating Window Native macOS Transparency, Bidirectional Millisecond Sync & Claude Defense] (Fixes #3733, Ref #3718, #3721, PR #3738, Thanks to @lbjlaq):
  • Native macOS Transparency & Window Drag Permissions: Added Tauri macos-private-api and native macOSPrivateApi: true configuration, eliminating WebKit black background edges; injected transparency scripts on the first frame and declared drag/always-on-top capabilities. (Fixes #3733, Ref #3718, PR #3738, Thanks to @lbjlaq)
  • Dark/Light Adaptation & Cross-Window Millisecond i18n Sync: Mini view dynamically inherits theme changes; bound app://theme-changed and app://language-changed cross-window events for bidirectional seamless synchronization across all 12 localization files. (Fixes #3733, PR #3738)
  • Claude Outbound Sentinel Purge & Thinking Budget Guard: Enforced defensive sentinel purging for Claude targets and clamped thinkingBudget >= 1024 to satisfy Anthropic upstream validation rules. (Ref #3721, PR #3738)
  • Unsigned Orphan Thought Graceful Degradation & Quota Card Accent Line: Gracefully degraded unsigned thoughts into plain text; refined GroupQuotaCard progress display into a subtle bottom accent line while preserving high-contrast satiety coloring. (PR #3738)
• [Official Model Suite Upgrade, Tiered Thinking Protocol & Large-Context QoS Scheduling] (Fixes #3735, Ref #3720):
  • Official Model Ingestion & Legacy Model Purging: Dynamically extracted officially supported models from fetchAvailableModels; purged hardcoded legacy models from configs and codebase, retaining only official flagship models and bare prefix derivatives. (Fixes #3735)
  • Standardized Context Window & Output Limits: Standardized context and output limits across OpenAI, Claude, and Gemini metadata endpoints (Gemini 4 Argon / 3.8 Flash: 1M / 65k, Claude 5.5: 1M / 128k, Claude 4.6: 200k / 64k, GPT-OSS: 131k / 32k). (Fixes #3735)
  • Tiered Thinking Protocol (Thinking Budget vs Thinking Level): Governed new-generation tiered models (Argon, Claude 5.5) exclusively with thinkingLevel and traditional models with thinkingBudget, eliminating HTTP 400 dual-field conflicts. (Fixes #3735)
  • Uncached Token Dynamic Allowance & Cold-Start Isolation: Tracked session context watermarks in QoSGuardian, applying a +7.0s latency allowance per 100K uncached tokens and raising TTFT thresholds to 18.5s. (Ref #3720)
  • One-Click QoS Jail Release & Remaining Cooldown Display: Displayed real-time QoS cooldowns on cards and tables, providing an instant manual release button and backend API (release_account_qos_jail). (Ref #3720)
• [Single-Language Catalog Purity & Navigation Synchronization]:
  • Navigation Structure & i18n Governance: Rebranded multi-protocol tab to "Model Catalog & Protocols" and aligned to the 3rd navigation slot, standardizing all 12 languages to clean, unpolluted native terms. (Ref #3733)
  • Quota Card Tooltip Full Localization: Fully localized GroupQuotaCard hover tooltips across all 12 dictionaries and removed redundant tooltips on capability badges.

• [上游 fetchAvailableModels 缺少 quotaInfo 架构级解耦与全系模型保活自适应] (Fixes #3737, Thanks to @xiaozhi-xmu, @chenshin166-arch):
  • 模型可用性与额度解析解耦: 彻底移除旧逻辑中 if let Some(quota_info) = info.quota_info 的强过滤门禁，以官方公开模型目录为基准，对 Google 上游不再为 Gemini 系（gemini-3.8-flash-*、gemini-3.1-flash-lite 等）下发 quotaInfo 的账号无损保留模型可用性，彻底消灭全系模型被误丢弃导致本地秒回 503 的严重缺陷。(Fixes #3737, Thanks to @xiaozhi-xmu, @chenshin166-arch)
  • 分组配额预拉取与短板木桶融合: 先行获取 retrieveUserQuotaSummary 分组配额（5h 与周配额双窗口），当模型缺失 quotaInfo 时直接从对应的 Gemini 组或第三方模型组中继承配额；Argon 优先匹配独立分组，无独立分组时优雅降级至通用 Gemini 分组。(Fixes #3737)
  • 严格保护显式零额度与最晚重置时间: 模型级明确为 0% 时绝不被正数分组额度覆盖；双窗口额度相同时优先采用较晚的重置时间以保护健康账号。(Fixes #3737)
  • 废弃内部代号 gemini-pro-agent 彻底清退: 在 variant_mapping.rs 中将 gemini-3.1-pro 家族规范映射由已废弃的内部代号 gemini-pro-agent 升级为官方标准物理模型 gemini-3.1-pro-high，彻底杜绝请求 3.1 Pro 时因调度器寻找已被基准线过滤淘汰的 gemini-pro-agent 导致 503 No accounts available with quota for model: gemini-pro-agent。(Fixes #3737)
  • Claude 协议补齐账号级动态模型自适应改写: 选号后调用 token_manager.resolve_dynamic_model_for_account，实现与 OpenAI 及 Gemini 协议处理逻辑严格对称，支持根据具体命中账号的实际授权目录自适应匹配候选物理模型。(Fixes #3737)
• [Gemini 4 Argon 全系列完整调度闭环与独立配额防穿透隔离] (Fixes #3735):
  • 调度器高阶模型与专属配额组能力继承对齐: 在 TokenManager::get_token_internal（候选池精炼）与 find_fallback_account_in_same_family 中，为候选账号过滤补齐 token_supports_advanced_model 与专属计费桶（normalized_target != "claude" && token.model_quotas.contains_key(...)）能力判定；彻底解决第一阶段收敛出具备 Argon 配额的账号后在第二阶段被 available_models 严格检查误杀、导致抛出 503 No accounts available with quota for model 的根本缺陷，同时严格保持 Claude 5.5 / 4.6 跨代隔离不串池。(Fixes #3735)
  • 全链路健壮识别配额组与能力持久化: 在 load_single_account、get_available_models_from_json、refresh_quota 以及 modules/quota.rs 中全面支持 camelCase 与 snake_case 双字段形态，只要分组或桶的 display_name 或 bucket_id 包含 argon 均健壮捕获其计费配额；当账号已具备 gemini-4-argon 时，自动双向补齐内存中的 available_models 与限额定义。(Fixes #3735)
  • 优化账号表格邮箱列排版与徽章防溢出: 重构 AccountTable.tsx 邮箱列布局为垂直弹性容器，将多徽章列表设为 flex-wrap 并移除 shrink-0，使 Argon、Claude 5.5、GPT-120B 等多标签在邮箱下方自然换行，杜绝横向穿透遮挡相邻“模型配额”列胶囊条。(Fixes #3735)
• [Mini 桌面悬浮视窗 macOS 原生真透明、双向毫秒级跨窗口同步与 Claude 拓扑防御增强] (Fixes #3733, Ref #3718, #3721, PR #3738, Thanks to @lbjlaq):
  • Mini 桌面悬浮视窗 macOS 原生真透明与拖拽权限健全: 引入 Tauri macos-private-api 特性与原生 macOSPrivateApi: true 配置，彻底消除 macOS WebKit 内核强制绘制底色导致的圆角黑边/白矩形；首帧注入透明脚本与 URL 标识，并在能力清单中补齐 mini 窗口拖拽及置顶操作权限。(Fixes #3733, Ref #3718, PR #3738, Thanks to @lbjlaq)
  • 深浅模式自适应与跨窗口多语言毫秒级同步: Mini 悬浮窗自适应继承主控制台与操作系统的暗黑/浅色主题；接入 app://theme-changed 与 app://language-changed 跨窗口事件监听，保证主控制台与 Mini 悬浮窗无论切换语言还是主题均毫秒级双向无感同步，并补全全部 12 国语言字典。(Fixes #3733, PR #3738)
  • Claude 出站门禁防御性哨兵清理与思考预算下限保护: 对目标为 Claude 模型的请求执行严格的防御性哨兵清理（Purge），彻底剥离历史会话或跨模型切换遗留的任何哨兵；出站阶段增加 Claude 思考预算下限动态保护（Clamp thinkingBudget >= 1024），防止因客户端默认值触发 Anthropic 上游 400 校验异常。(Ref #3721, PR #3738)
  • 无签名孤儿思考块安全降级与配额卡片底部细线进度点缀: 跨模型切换或历史无合法 Claude 签名时，安全丢弃思考块或平滑降级为普通正文；将 GroupQuotaCard 进度条融合为底部精致细线点缀，兼顾多徽章与动态标签展示，并保留满额度高对比度变色逻辑。(PR #3738)
• [官方权威模型全景升级、分级思考传输与大上下文自适应 QoS 调度] (Fixes #3735, Ref #3720):
  • 官方权威模型提取与过时模型全面下线: 基于官方 fetchAvailableModels（agentModelSorts 推荐列表、imageGenerationModelIds、tieredModelIds）动态提取每个账号真实支持的旗舰模型；在代码库与配置文件（model_specs.json, model_mapping.rs, cli_sync.rs, opencode_sync.rs, modelConfig.ts 等）中彻底清除历史遗留的硬编码过时模型（如 gemini-2.0/2.5、claude-3-5/3-7、chat_*、tab_* 等），仅保留官方权威模型及后缀派生裸模型。(Fixes #3735)
  • 全协议端点上下文窗口与最大输出参数标准化: 在 OpenAI（/v1/models）、Claude（/v1/models）与 Gemini（/v1beta/models）元数据端点中引入并标准化 context_length、max_tokens、max_output_tokens 等字段；严格依据官方规格动态返回真实限制（Gemini 4 Argon / 3.8 Flash 为 1,048,576 / 65,536，Claude 5.5 为 1,000,000 / 128,000，Claude 4.6 为 200,000 / 64,000，GPT-OSS 为 131,072 / 32,768）。(Fixes #3735)
  • 思考预算与思考等级分类模型传输 (Thinking Budget vs Thinking Level): 新一代分级模型（Gemini 4 Argon、Claude 5.5 系列）受控于 thinkingLevel（"LOW", "MEDIUM", "HIGH"），严格下发思考等级，严禁输出数字预算；传统预算模型（Gemini 3.8 Flash、Gemini 3.1 Pro、Claude 4.6 等）走 thinkingBudget 通道，杜绝双字段冲突 400 报错。(Fixes #3735)
  • 未缓存 Token 动态耗时补偿与冷启动隔离: 在 QoSGuardian 中引入会话上下文水位跟踪，自动识别超长上下文（如 340k+ 零缓存）冷启动 Prefill 阶段，按每 100K 未缓存 Token 补偿 +7.0s 判定阈值，并放宽平反首字基准阈值至 18.5s。(Ref #3720)
  • QoS 小黑屋一键解除与状态倒计时透明化: 账号管理卡片与表格视图展示 QoS 小黑屋实时剩余时长，并提供一键解除小黑屋限制按钮及后台接口（release_account_qos_jail）。(Ref #3720)
• [多语言单语纯粹性治理与导航排版对齐]:
  • 导航结构与多语言规范: 将多协议支持更新为“模型大全与协议”并调换至第三个导航位，12 种系统语言全面对齐单语纯粹性，清除中英混杂。(Ref #3733)
  • 配额卡片与提示全量国际化: GroupQuotaCard 悬浮提示卡片完整接入多语言字典，并清理能力徽章上的冗余悬停文案。

What's Changed

• feat(ui,proxy): Mini 桌面悬浮视窗圆角真透明、跨窗口同步与 Claude 拓扑防御增强 (Fixes #3733, Ref #3718, #3721) by @lbjlaq in #3738

Full Changelog: v4.9.9-beta.10...v5.0.0-beta.0