v2.1.20

✨ 本次更新

• 优化了响应截断的处理逻辑，并修复了流式请求失败时错误返回 HTTP 200 的问题，确保客户端能正确识别错误并触发重试。
• 显著提升了不同模型供应商（如 OpenAI、Anthropic、Gemini）之间的兼容性，修复了工具调用、图片上传及系统消息处理中的多项问题。
• 增强了会话稳定性，修复了连接中断后可能导致会话永久卡死在 502 错误状态的问题。
• 修复了界面显示问题，现在能准确统计并显示 Antigravity 账号的状态与数量。
• 优化了代理转发逻辑，自动绕过本地回环地址（localhost），解决了开启代理时无法连接本地模型的问题。
• 修复了图片生成接口（/v1/images/generations）被错误屏蔽的问题，恢复了账号模式下的生图功能。
• 解决了 Docker 部署中的多项问题，包括特定系统下的启动崩溃、配置卷文件无法自动更新以及计费表加载失败等。

What's New

• Improved handling of truncated responses and fixed an issue where streaming errors were incorrectly reported as successful (HTTP 200), allowing clients to correctly trigger retries.
• Enhanced compatibility between different AI providers (OpenAI, Anthropic, Gemini), fixing multiple issues related to tool calling, image uploads, and system message handling.
• Improved session resilience by fixing a bug where interrupted connections could cause a session to get stuck in a persistent 502 error state.
• Fixed UI bugs to ensure Antigravity accounts are correctly counted and displayed in the account list and statistics.
• Optimized proxy routing to automatically bypass local addresses (localhost), resolving connection failures to local models when a system proxy is active.
• Restored the image generation API (/v1/images/generations) functionality which was previously being incorrectly blocked.
• Resolved several Docker deployment issues, including startup crashes on certain systems, missing configuration files after updates, and pricing table loading failures.

完整提交记录 / Full commit list

• fix(translation): report truncation from response.incomplete as max_tokens/refusal