HB00/dsh-llm-failover

README 摘要

dsh-llm-failover Provider failover for DeepSeek Harness ( dsh ): when a model provider returns rate-limit (429) or quota-exhausted errors, switch to the next provider automatically — with a configurable per-provider model mapping, a permanent final fallback, and a visible notice in the UI. 模型提供方故障转移插件：遇到限流（429）或配额耗尽时，自动切换到下一个提供方——支持按提供方配置模型、永久最终兜底，并在界面显示切换提示。 Features / 功能 - Automatic switch on RATE LIMIT / QUOTA failures, with configurable retry threshold ( fallbackAfterRetries ). / 遇到限流/配额错误自动切换，可配置切换前连续失败次数。 - Per-provider model mapping — a switched provider can use a different model. / 每个提供方可单独指定切换后使用的模型。 - Cooldown per provider ( cooldownMs ): a cooled-down provider is skipped until it recovers. / 提供方冷却机制：冷却期间直接跳过，到期自动恢复。 - The last entry in the provider list is the permanent fallback and is never switched away (no retry loops). / 列表最后一条是永久兜底，永远不会被切走（不会死循环）。 - UI notice bar on every switch (auto-dismiss). / 每次切换在输入框上方显示提示条（自动消失）。 - Configuration card in Settings → Plugins → Plugin configuration. / 设置 → 插件 → 插件配置 页面卡片直接配置。 Install / 安装 Configuration / 配置 Either use the Settings UI card, or edit /.dsh/settings.yaml : How it works / 原理 Hooks two official agent waterfalls: - agent…

分类