geekchongv/ccswitch-context-guard

CCProxy Agent is a small local safety layer that sits between the Claude Code or Claude Desktop AI coding tools and another tool called ccswitch, which routes requests to different underlying AI models. It exists to solve one specific, annoying problem: AI models have a maximum amount of text they can handle in one conversation, called a context window, and when a session gets too long or a request asks for too much output, the model returns an error and the whole session can crash, losing your work. CCProxy Agent quietly checks the size of each outgoing request before it leaves your machine and lowers the requested output length so the total, combined with what you already typed, stays under the model's limit. If an error still slips through, it reads the details from that error message, recalculates a safer limit, and automatically retries the request once instead of making you restart. It also handles a few related problems, such as trimming old tool results that are no longer needed, warning you when a conversation is getting close to its limit, and converting images sent to Claude Desktop into a text description so models that cannot see images can still respond sensibly. You can run it as a ready made Windows executable or a macOS app, both of which create their own local settings file the first time you start them, or you can run it from source code using Node.js. Once running, it briefly changes your Claude CLI or Claude Desktop settings to point at itself, and restores your original settings automatically when you close it. It also includes a small built in web dashboard on your own computer where you can see its status, adjust settings, and watch live logs of what it is doing. This is an independent, community built project, not an official tool from Anthropic or the ccswitch project. It is aimed at developers who already use Claude Code or Claude Desktop with ccswitch and have run into context length crashes, and setup requires comfort with running a background application or, for the source install, basic command line and Node.js use.