Bug Fixes
- Fixed reasoning items duplicating their text and a crash when the summary field was missing. (#2855)
- Fixed replies held by another filter not going through reasoning cleanup. (#2855)
- Fixed reasoning text being lost when passed to another model; it is now retained as a summary. (#2855)
- Fixed rate limit reset timing being measured from an in-flight request instead of the last successful response. (#2718)
- Fixed vendor-provided rate limit cooldowns not being honored until they end and incorrectly restoring account/image picker state. (#2718)
- Fixed a shorter failure shortening a longer rate limit cooldown. (#3289)
- Fixed launcher detection in VS Code 1.139 when using the versioned layout.
- Fixed vendor model aliases being counted as model swaps. (#4593)
Improvements
- The gateway now uses vendor-provided quota reset times instead of estimating them. (#3289)
- When a vendor reports that a model is not served in the current region, requests now fail over to the next model in the group. (#2887)
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>
Bug Fixes
- 修复了推理项重复其文本以及摘要字段缺失时崩溃的问题。(#2855)
- 修复了被另一个过滤器保留的回复未经过推理清理的问题。(#2855)
- 修复了推理文本传递给另一个模型时丢失的问题；现在会将其保留为摘要。(#2855)
- 修复了速率限制重置计时从进行中的请求而非最后一次成功响应开始计算的问题。(#2718)
- 修复了供应商提供的速率限制冷却时间在结束前未被遵守，以及错误恢复账户/图像选择器状态的问题。(#2718)
- 修复了较短的故障缩短较长速率限制冷却时间的问题。(#3289)
- 修复了使用版本化布局时 VS Code 1.139 中的启动器检测问题。
- 修复了供应商模型别名被计为模型切换的问题。(#4593)
Improvements
- 网关现在使用供应商提供的配额重置时间，而不是估算它们。(#3289)
- 当供应商报告某模型不在当前区域提供服务时，请求现在会故障转移到组中的下一个模型。(#2887)
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。