lbjlaq/Antigravity-Manager v4.9.8-beta.5

Antigravity Tools v4.9.8-beta.5

on GitHub

• [Switch Quota & Model Discovery UA to Official Hub Identity with Dynamic Versioning] (Ref #3665, Thanks to @wd2es7):
  • Switch Upstream Discovery UA to Hub Subclient: Fixed an issue where ULTRA accounts could not discover or route gemini-4-argon models. Following community insights from @wd2es7 and reverse engineering of /Applications/Antigravity.app, verified that Google gates Argon availability based on the client subclient_type in the User-Agent (ide/vscode filters out Argon yielding 35 models, whereas hub/cli returns all 38 models including Argon). Consolidated fetch_project_id, fetch_quota_with_cache, and fetch_quota_summary to official antigravity/hub UA. (Ref #3665, Thanks to @wd2es7)
  • Dynamic Versioning & OS Fingerprint: Eliminated hardcoded version 2.17.0 in constants::USER_AGENT, replacing it with build_official_upstream_user_agent() which dynamically embeds the current app version (e.g. 4.9.8-beta.5) and host architecture/platform (darwin/linux/windows and arm64/x64), ensuring upstream headers stay synchronized with releases. (Ref #3665)
  • Official Model Catalog Alignment: Registered gemini-4-argon in offline fallback official_models.json aligned with upstream minThinkingBudget: 32 metadata. (Ref #3665)
• [Gemini 4 Argon Runtime Thinking Budget Safeguard & 400 Error Mitigation] (Ref #3665):
  • Runtime Minimum Thinking Budget Guard (≥ 1024): Upstream verification revealed that despite Google's metadata indicating minThinkingBudget: 32, sending any non-zero thinking budget $< 1024$ to generateContent triggers 400 Thinking level is unsupported: THINKING_LEVEL_MINIMAL. Added an inbound pipeline clamp in inbound.rs and aligned gemini-4-argon-low in model_specs.rs to enforce a runtime floor of 1024, completely preventing upstream 400 failures. (Ref #3665)

• [配额与模型发现 UA 切换为 Hub 官方标识与动态版本号注入] (Ref #3665, Thanks to @wd2es7):
  • 配额刷新与模型探活 UA 全面切换为 Hub 标识: 解决 ULTRA 账号无法探知并调度 gemini-4-argon 系列模型的问题。根据社区贡献者 (@wd2es7) 定位线索并经官方 Antigravity 桌面端二进制逆向及真实上游抓包验证，Google 服务端依客户端 subclient_type 执行白名单模型门禁（ide/vscode 过滤 Argon 仅下发 35 个模型，而 hub/cli 下发包含 Argon 在内的 38 个完整模型目录）。将 quota.rs 中的 fetch_project_id、fetch_quota_with_cache 及 fetch_quota_summary 的请求头统一收敛为官方 antigravity/hub 规范。 (Ref #3665, Thanks to @wd2es7)
  • 动态集成应用版本与操作系统指纹: 废除 constants::USER_AGENT 中硬编码的 2.17.0 版本号，重构为 build_official_upstream_user_agent() 动态读取当前项目实时版本（如 4.9.8-beta.5）及运行平台（darwin/linux/windows 及 arm64/x64），确保上游请求标识与发布版本严格原子同步。 (Ref #3665)
  • 官方模型目录补齐与下限对齐: 在内置模型兜底表 official_models.json 中补齐 gemini-4-argon 官方定义，对齐官方元数据标称的 minThinkingBudget: 32。 (Ref #3665)
• [Gemini 4 Argon 运行时思考预算下限保护与 400 报错防护] (Ref #3665):
  • 执行层思考预算下限硬保护 (≥ 1024): 经上游真实请求验证，Google 官方对 Gemini 4 (Argon) 施加了未在元数据中公开的运行期限制——当 thinkingBudget < 1024（且非 0）时，上游接口将直接抛出 400 Thinking level is unsupported: THINKING_LEVEL_MINIMAL。为此，在管线入站 inbound.rs 与档位计算 model_specs.rs 中将 gemini-4-argon-low 及客户端传入的低预算安全钳位在 1024，彻底规避上游 400 崩溃。 (Ref #3665)

Full Changelog: v4.9.8-beta.4...v4.9.8-beta.5