v1.0.62

MODELSWAP v1.0.62

中文

文档站与 README 的 SEO/GEO 基建：搜索引擎与 AI 助手更容易发现并准确引用 ModelSwap。

新增

• docs.modelswap.app 构建时自动产出 robots.txt、sitemap.xml（15 章 × 中英双语 hreflang）与 llms.txt，每章新增 canonical 与 OG 标签。
• README 新增「支持的 Agent」清单（10 个适配器全列名），并链接 modelswap.app/llms.txt 机器可读产品档案。

修复

• 修正 README 过时描述：移除已下线的「项目绑定注入 .env」说法，自动创建平台数统一为 31（浏览器扩展）。

English

SEO/GEO groundwork for the docs site and READMEs: search engines and AI assistants can discover and accurately cite ModelSwap.

New

• docs.modelswap.app now emits robots.txt, sitemap.xml (15 chapters × zh/en with hreflang) and llms.txt at build time, plus per-chapter canonical and OG tags.
• The READMEs gain a Supported agents section listing all 10 adapters by name, plus links to the machine-readable product reference at modelswap.app/llms.txt.

Fixed

• Stale README claims corrected: the removed project-binding .env injection is gone, and the auto-create platform count is unified at 31 (browser extension).

v1.0.61

MODELSWAP v1.0.61

中文

用户手册全面重写：新增 ModelSwap 介绍与十分钟上手，六组侧边栏大纲，各功能章节按现有界面补全细节与截图。

改进

• 手册新增「ModelSwap 是什么」介绍章与「十分钟上手」快速上手章，先了解再安装，跑通第一次模型切换。
• Agent 配置、模型管控、用量统计、密钥库等章节按现有界面补全：添加站点、查看配置文件、连接测试自动拉模型、用量告警阈值与查询凭证配置等均有分步说明。
• 手册补齐桌面版优先的安装指引（含 macOS 首次启动放行步骤）与局域网同步配对全流程，并新增多张功能截图。

English

The user manual gets a full rewrite: a new ModelSwap intro and ten-minute quick start, a six-group sidebar outline, and feature chapters completed with current-UI details and screenshots.

Improved

• Two new opening chapters — 'What is ModelSwap' and a ten-minute Quick Start — so you understand the tool before installing and finish your first model switch in minutes.
• Feature chapters now match the current UI step by step — adding sites, viewing agent config files, connection tests that auto-fetch models, usage alert thresholds, and balance-credential setup.
• Desktop-first install guidance (including the macOS first-launch allow steps), a complete LAN sync pairing walkthrough, and several new feature screenshots.

v1.0.60

MODELSWAP v1.0.60

中文

清理了早期的项目绑定功能：不再有 .modelswapenv 文件和相关命令，密钥管理聚焦在存储、终端注入和多机同步上。

改进

• 移除了 vault env 等项目绑定命令：生成 .env、登记关联、按项目刷新的整套流程不再提供，界面和文档也不再提及。
• vault inject 改为显式指定密钥：用 --keys 列出要注入的密钥名，不再读取项目目录里的 .modelswapenv 文件。
• 随产品分发的 Agent Skill 与用户手册同步更新，不再引导使用已移除的命令。

English

The early project-binding feature is removed: no more .modelswapenv files or related commands. Key management now focuses on storage, terminal injection, and multi-device sync.

Improved

• Project-binding commands such as vault env are removed — .env generation, binding registration, and per-project refresh are gone from the CLI, UI, and docs.
• vault inject now takes an explicit --keys list and no longer reads a .modelswapenv file from your project.
• The bundled Agent Skill and the user manual are updated accordingly and no longer reference removed commands.

v1.0.59

MODELSWAP v1.0.59

中文

更换或换绑密钥后，各 Agent 的配置会自动用上新密钥，不再需要手动重新切换站点。

改进

• 在密钥管理里修改某个密钥的值后，所有用到它的 Agent 配置自动同步更新。
• 站点换绑其他密钥（或修改接口地址）后，相关 Agent 配置同样自动更新。
• 命令行执行 modelswap vault set 保存密钥后，会提示已同步更新了几个 Agent 配置。

English

After changing or rebinding a key, agent configs pick up the new key automatically — no manual site re-switching needed.

Improved

• Editing a key's value in the vault now automatically updates every agent config that uses it.
• Rebinding a site to a different key (or changing its endpoint) also refreshes the affected agent configs automatically.
• The modelswap vault set command now reports how many agent configs were updated with the new key.

v1.0.58

MODELSWAP v1.0.58

中文

更新体验优化：不再重复弹出更新提示，更新弹框排版更清爽，更新说明改用大白话。

改进

• 更新弹框排版优化：移除标题旁的下载图标，「新增 / 改进 / 修复」分类改为醒目的小标题。
• 更新说明改用大白话：每条只说这一版做了什么，技术细节不再出现在更新内容里。

修复

• 修复更新提示重复弹出的问题：打开更新弹框后，不再额外弹出一条重复的提示条。

English

A cleaner update experience: no more duplicate update notices, a tidier update card, and plain-language release notes.

Improved

• The update card gets a cleaner layout: the decorative download icon is gone and the New / Improved / Fixed labels are now proper section headings.
• Release notes are rewritten in plain language: each item states what changed, without technical detail.

Fixed

• Fixed duplicate update notices: opening the update card no longer triggers a second redundant tooltip.

v1.0.57

MODELSWAP v1.0.57

中文

自动创建密钥的平台列表从 33 个精简到 31 个；Cloudflare 和 Google AI Studio 改为手动添加密钥，其余功能不变。

改进

• 自动创建列表移除了 Cloudflare 和 Google AI Studio 两个入口，保留 31 个一键创建的平台。
• 这两个平台仍可正常使用：手动添加密钥即可，模型和同步功能不受影响。
• 文档与界面中的平台数量说明统一更新为 31。

English

The auto-create key list is streamlined from 33 platforms to 31; Cloudflare and Google AI Studio keys are now added manually. Everything else is unchanged.

Improved

• Cloudflare and Google AI Studio entries are removed from the auto-create list; 31 one-click platforms remain.
• Both platforms still work: add their keys manually — models and sync are unaffected.
• Docs and UI copy now consistently say 31 platforms.

v1.0.56

MODELSWAP v1.0.56

中文

修复 CLI 切换站点时把平台全量模型清单写进客户端的问题：此前 modelswap provider use/switch 会把站点的全部模型（例如 Google 的 55 个）镜像进 ZCode 等客户端的模型选择器，而你在 ModelSwap 里明明只勾选了一个。现在 CLI 与网页端行为一致——只镜像你选中的模型。

改进

• ZCode 模型选择器不再被长列表淹没：站点勾选几个就显示几个，与 ModelSwap 界面上的勾选一一对应。

修复

• CLI 切换（provider use / provider switch）不再传全量模型清单：只把你切换到的模型（以及 Claude 档位映射里的模型）写进客户端配置，网页端一直是这个行为，CLI 此前保留了大重构前的旧逻辑。
• 修复 Windows CI 上 provider 生命周期测试偶发的 EBUSY 文件锁（排队中的配置原子写入与测试裸读 user.json 竞态），此前该偶发问题曾两次阻塞发版流水线。

English

Fixes the CLI flooding agent model pickers with the provider's full catalog: modelswap provider use/switch used to mirror every discovered model (e.g. all 55 from Google) into clients like ZCode even when only one was selected in ModelSwap. The CLI now matches the Web behavior — only your selected models are mirrored.

Improved

• ZCode's model picker is no longer flooded: it shows exactly the models enabled for the site, one-to-one with your ModelSwap selections.

Fixed

• CLI switches (provider use / provider switch) no longer pass the full model catalog: only the switched-to model (plus Claude tier mappings) is written to agent configs — the Web path always behaved this way; the CLI had kept the pre-refactor behavior.
• Fixes an intermittent EBUSY file lock in the provider lifecycle test on Windows CI (queued atomic config writes racing the test's raw user.json reads); the flake had blocked the release pipeline twice.

v1.0.55

MODELSWAP v1.0.55

中文

修复 Google AI Studio 模型名带 models/ 前缀的问题：Gemini 的模型列表接口返回的是资源名（如 models/gemini-2.5-flash），此前被原样导入，导致 ZCode 等客户端模型选择器里出现一长串带前缀的名字。现在导入时统一剥离前缀，存量数据也会自动修复。

改进

• 附带收益：剥离前缀后模型 id 与 models.dev 目录键一致，Google 模型现在能正确挂上上下文长度/输出上限等元数据。

修复

• 模型导入时剥离 Google 的 models/ 资源名前缀（Gemini API 两种写法都接受，剥前缀后 ZCode 模型选择器显示干净的 gemini-3.8-flash 等短名）。
• 存量数据自愈：models-cache.json 里已缓存的带前缀模型、user.json 里各 Agent 的模型选择（含 activeModelId 与模型覆盖配置）在读取时自动剥前缀并持久化，无需手动清理。

English

Fixes Google AI Studio model ids carrying the models/ prefix: Gemini's model directory returns resource names (e.g. models/gemini-2.5-flash) which were previously imported verbatim, leaving long prefixed names in agent model pickers such as ZCode's. Imports now strip the prefix, and existing local data is healed automatically.

Improved

• Side benefit: with bare ids matching models.dev catalog keys, Google models now correctly pick up context/output-limit metadata during enrichment.

Fixed

• Model discovery now strips Google's models/ resource-name prefix (the Gemini API accepts both forms; after the fix pickers show clean short names like gemini-3.8-flash).
• Existing data heals itself: cached models and every agent's model selections (including activeModelId and model overrides) are stripped and persisted on the next load — no manual cleanup needed.

v1.0.54

MODELSWAP v1.0.54

中文

gcloud 一键创建 Google AI Studio Key 的可靠性加固：建 key 后立即真实调用验证，失败自动换项目重试。起因是实测发现 Google 会对全新且未绑定付款方式的项目自动封禁生成请求（403 project denied），而复用已有项目稳定可用。

改进

• 项目选择策略：优先复用已有 AI Studio/ModelSwap 命名项目（Google 信任度高），新建仅作兜底；绝不触碰账号下无关项目。

修复

• 新建 key 后立即用最小生成调用验证可用性（models 列表不足以暴露项目封禁），失败即删除坏 key 并按候选项目顺序重试。
• 全部候选失败时的报错附操作指引（先在 AI Studio 网页生成一次 Key 建立可信项目，再回来复用）。

English

Hardens gcloud key creation for Google AI Studio: every freshly created key is verified with a real generation call, and failures automatically retry on another project. This follows a live finding — Google auto-denies generation on brand-new no-billing projects (403 project denied) while existing projects keep working.

Improved

• Project selection now prefers existing AI-Studio-style/ModelSwap projects (most trusted by Google); creating a new one is the fallback, and unrelated account projects are never touched.

Fixed

• New keys are verified with a minimal generation call (model listing cannot surface a project-level deny); on failure the bad key is deleted and the next candidate project is retried.
• When every candidate fails, the error now includes actionable guidance (create one key in the AI Studio web console first to establish a trusted project, then reuse it here).

v1.0.53

MODELSWAP v1.0.53

中文

补上 Google AI Studio 的平台图标：采用官方 Gemini 四角星渐变标，模型页/首页/用量卡等处的 Google 条目不再回退为字母头像。

新增

• 新增 Google（Gemini）官方四角星图标，正方形矢量，注册进 PROVIDER_ICON 映射。

改进

• 图标为矢量 SVG，明暗主题均正常显示（彩色渐变无需反色处理）。

修复

• v1.0.51 引入 Google 预设时遗漏了图标映射，本次补齐。

English

Adds the missing Google AI Studio provider icon: the official Gemini sparkle in gradient, so Google entries on the models page, dashboard, and usage cards no longer fall back to a letter avatar.

New

• New square vector icon built from the official Gemini sparkle, registered in the PROVIDER_ICON map.

Improved

• The icon is vector SVG and renders correctly in both themes (colorful gradient needs no dark-mode inversion).

Fixed

• The Google preset shipped in v1.0.51 without an icon mapping; this closes the gap.