[BUG] Intermittent Network Error during DeepSeek "Thinking" phase when MCP/WebSearch is enabled (M4 Mac Mini / Sequoia)

[motasimmasood]

Bug Description

I am experiencing intermittent Network Error messages when using DeepSeek models in Chatbox, but specifically only when MCP or WebSearch tools are enabled.
The Pattern:

• The error occurs during the "Thinking" (reasoning) phase, usually between 5 and 120 seconds into the generation.
• It often happens when the model is attempting to fetch information via an MCP tool or right after a tool has returned data. But can also happen during just thinking output but less often.
• It is not constant; it happens intermittently, suggesting a race condition or a timeout issue during high-concurrency tasks.
• Hardware specific: This setup (same API key, same network, same MCP config) works perfectly on a MacBook Pro, but fails on an M4 Mac Mini running macOS Sequoia.

Steps to Reproduce

1. Configure DeepSeek with Thinking/Reasoning enabled.
2. Enable any MCP tool or the built-in Web Search.
3. Send a complex query requiring both reasoning and a tool call.
4. Watch the thinking phase. The error typically triggers while the "Thinking" bubble is active or when the tool is transitioning back to text generation.

Expected Results
The connection should remain open regardless of how long the "Thinking" or "Tool Call" phase takes (up to the API's maximum timeout).

Actual Results
Network Error only during thinking phase, most commonly when thinking calls MCP tool

Screenshots
If possible, please add screenshots to help explain the issue.

Desktop (please complete the following information):

OS: macOS Sequoia (15.x)
Hardware: Mac Mini M4
Chatbox Version: Version 1.20.1 (1.20.1)
Model: DeepSeek (Official API)
Tooling: MCP (Local/Remote) and/or Web Search (Tavily)

Additional Context
Troubleshooting Steps Taken: The following attempts were made to stabilize the connection on the M4 Mini:

1. Network Configuration: Disabled "Limit IP Address Tracking".
2. Latency Monitoring: Observed that pings to api.deepseek.com remain stable until 5-120 seconds into the thinking phase, then suddenly jitter increases from 11-20 seconds all the way to 100+ seconds, likely when MCP tool is called; causing Chatbox to drop the primary API socket.
3. Hardware Comparison: Verified the identical setup on a MacBook Pro (Intel based) works without any Network Error, pointing to a potential issue with the M4's networking stack or Sequoia's energy management of background processes in Chatbox.
4. Log Analysis: macOS log stream shows RunningBoard and LaunchServices performing frequent inheritance/permission checks on Chatbox during the exact window the error occurs. It appears the OS might be "throttling" the app's priority while it waits for a long-running tool call to return.

---

Bug 描述
清晰简洁地描述这个 bug 是什么。

重现步骤
请提供能够让我们重现这个 bug 的步骤：

1. 前往 "......"
2. 点击 "......"
3. 滚动到 "......"
4. 发现了这个 bug。

期望结果
请清晰简洁地描述预期的行为。

实际结果
请清晰简洁地描述实际的行为。

截图
如果可行，添加截图以帮助解释问题。

桌面端（请填写以下信息）：

• 操作系统：[例如 macOS]
• 应用程序版本：[例如 2.0.1]

其他上下文
在这里提供关于问题的任何其他上下文，例如与其他软件或应用程序的交互等。