yetone/magpie v0.1.1224

magpie v0.1.1224

on GitHub

Features

• Each WorkBuddy, Trae CN, MiniMax, Qoder, or plugin account can now have its daily check-in turned off individually.
• Request logs now show the service tier requested, the tier used on each try, and the tier reported by the vendor.
• Library → Instructions now includes Qwen Code and writes its ~/.qwen/QWEN.md.
• Ollama, LM Studio, and oMLX now accept custom server addresses, so a preset can point to a server on another port or computer.
• Groups can now route by weight, with each member’s share shown next to it.
• Grok Build installed via Homebrew’s grok-build cask is now detected, shows its version, and updates with Homebrew.
• Renamed or moved Claude Code project folders can now be relocated together with their sessions from the Sessions page or CLI. The relocation UI is localized in zh-TW, ja, and de.
• Enterprise WorkBuddy accounts now show credits from the enterprise meter on the plan card.
• Ollama Cloud legacy plans now show session and weekly limits on the plan card.
• omp sessions in a WSL distro are now listed on Windows in the Sessions page and Usage → Sessions.

Bug Fixes

• Fixed Codex account switching so it changes only its own sign-in between accounts, not through the gateway.
• Fixed a fast model switch losing the client’s own Ultrafast request.
• Fixed conversations failing to compact after the search tool is removed on Responses vendors that don’t support that tool call type.
• Fixed Antigravity’s retired 2.5 Flash models being mislabeled as “Gemini 3.5 Flash Lite.”
• Fixed masked words being hidden inside tool definitions or schemas, so Codex’s patch tool instructions reach OpenAI intact.
• Fixed OpenCode Zen’s free models not answering.
• Fixed GLM Coding Plan keys at the pay-as-you-go endpoint not being retried at the plan endpoint; the provider now uses the plan endpoint.
• Fixed multi-protocol providers losing some endpoints’ models in the catalog.
• Fixed Homebrew-installed agents (gemini-cli, pi, cline) updating with npm instead of Homebrew.
• Fixed a Claude Code test or /usage run not ending with the process it started after magpie gave up, including npm’s node on Windows.
• Fixed picking a magpie model for omp so its picker and Ctrl+P now list every model magpie provides.
• Fixed Claude Desktop’s built-in skills showing as five failed writes instead of one row on the Skills tab.
• Fixed plugin check-in overrides for Trae, WorkBuddy, MiniMax, and Qoder, plus saved auto check-in, resetting after a state refresh.
• Fixed a half-written autostart launch agent being read as “Open at Login.”
• Fixed providers keeping queue, limits, and routing for only the first account; every account now keeps its own.
• Fixed Export Markdown overlapping the taller folder row.

Improvements

• The RTK page and Installed plugin list now open without waiting for the package manager or plugin vendors.
• oMLX’s decision API now follows its Server address setting.
• Heatmaps with a few busy days now shade each day by its share of the busiest.
• Model naming preferences now apply consistently to connected agents and to Qwen and Claude magpie models.

Install

Download from usemagpie.ai, or curl -fsSL https://usemagpie.ai/install.sh | sh. An installed magpie updates itself.

• macOS (Apple Silicon): magpie-darwin-arm64.dmg
• macOS (Intel): magpie-darwin-amd64.dmg
• Windows: magpie-windows-amd64.exe (ARM: magpie-windows-arm64.exe)
• Linux: magpie-linux-amd64 / magpie-linux-arm64 (needs GTK 3 and WebKitGTK 4.1)
• Terminal only: magpie-cli-<os>-<arch>

Features

• 每个 WorkBuddy、Trae CN、MiniMax、Qoder 或插件账户现在都可以单独关闭每日签到。
• 请求日志现在会显示所请求的服务层级、每次尝试实际使用的层级，以及供应商报告的服务层级。
• 资料库 → 指令现在包含 Qwen Code，并会写入其 ~/.qwen/QWEN.md。
• Ollama、LM Studio 和 oMLX 现在接受自定义服务器地址，因此预设可以指向其他端口或计算机上的服务器。
• 分组现在可以按权重路由，每个成员的份额显示在其旁边。
• 现在会检测到通过 Homebrew grok-build cask 安装的 Grok Build，显示其版本，并随 Homebrew 更新。
• 重命名或移动后的 Claude Code 项目文件夹现在可以从会话页面或 CLI 连同其会话一起重新定位。重新定位界面已本地化为 zh-TW、ja 和 de。
• 企业版 WorkBuddy 账户现在会在套餐卡片上显示来自企业计量表的积分。
• Ollama Cloud 旧版套餐现在会在套餐卡片上显示会话限制和每周限制。
• WSL 发行版中的 omp 会话现在会列在 Windows 上的会话页面和使用情况 → 会话中。

Bug Fixes

• 修复了 Codex 账户切换，使其只在自身账户之间更改登录，而不会通过网关。
• 修复了快速切换模型时丢失客户端自身 Ultrafast 请求的问题。
• 修复了在不支持该工具调用类型的 Responses 供应商上移除搜索工具后，对话压缩失败的问题。
• 修复了 Antigravity 已退役的 2.5 Flash 模型被错误标记为“Gemini 3.5 Flash Lite”的问题。
• 修复了被屏蔽的词语在工具定义或 schema 内被隐藏的问题，使 Codex 的 patch 工具指令能够完整原样地到达 OpenAI。
• 修复了 OpenCode Zen 免费模型不回复的问题。
• 修复了 GLM Coding Plan 在按量付费端点的密钥不会在套餐端点重试的问题；该供应商现在使用套餐端点。
• 修复了多协议供应商在目录中丢失某些端点的模型的问题。
• 修复了通过 Homebrew 安装的代理（gemini-cli、pi、cline）使用 npm 而不是 Homebrew 更新的问题。
• 修复了在 magpie 放弃后，Claude Code 测试或 /usage 运行未将其启动的进程一并结束的问题，包括 Windows 上 npm 的 node 进程。
• 修复了为 omp 选择 magpie 模型的问题，现在其选择器和 Ctrl+P 会列出 magpie 提供的所有模型。
• 修复了 Claude Desktop 的内置技能在技能标签页中显示为五次失败写入，而不是一行的问题。
• 修复了 Trae、WorkBuddy、MiniMax 和 Qoder 的插件签到覆盖设置，以及已保存的自动签到，在状态刷新后重置的问题。
• 修复了半写入的自动启动 launch agent 被识别为“Open at Login”的问题。
• 修复了供应商只为第一个账户保留队列、限制和路由的问题；现在每个账户都会保留自己的。
• 修复了 Export Markdown 与较高的文件夹行重叠的问题。

Improvements

• RTK 页面和已安装插件列表现在打开时无需等待包管理器或插件供应商。
• oMLX 的决策 API 现在会遵循其服务器地址设置。
• 对于只有少数繁忙日期的热力图，现在会按各日期占最繁忙日期的比例着色。
• 模型命名偏好现在会一致地应用到已连接的代理以及 Qwen 和 Claude 的 magpie 模型。

安装

从 usemagpie.ai 下载，或运行 curl -fsSL https://usemagpie.ai/install.sh | sh。已安装的 magpie 会自动更新。

The same release, and the feed magpie updates from: https://github.com/yetone/magpie-releases/releases/tag/v0.1.1224