GPT AntiCurse 作者： mistrjirka
Speed up very long ChatGPT conversations, load older history on demand, and export a chat to continue the work in a new conversation.
实验性实验性
扩展元数据
屏幕截图
关于此扩展
GPT AntiCurse is made for people who keep the same ChatGPT conversation running for a long time.
Large coding, research, debugging, and agentic-work chats can eventually contain hundreds or thousands of turns plus tool calls, hidden records, branches, and other internal conversation state. Even when the model is idle, processing all of that state can make the ChatGPT tab noticeably slower.
AntiCurse keeps a smaller recent part of the conversation active in ChatGPT instead of making the frontend process the complete historical graph every time.
There are two history modes.
Recent N + button keeps a configurable recent window in ChatGPT's normal interface. When you need older messages, a Load previous button appears above the conversation.
Auto window uses the same bounded recent history, but loads older archived messages automatically as you scroll toward the beginning of the chat.
Older history is kept outside ChatGPT's main React conversation state and rendered using a lightweight, virtualized view. Only a small number of archived pages stay mounted at once, so loading more history does not simply recreate the original performance problem.
AntiCurse can also keep a local backup of the conversation and export it to Markdown.
This is particularly useful for long-running projects. If a conversation eventually becomes too large for the model's useful context, you can export it, open a fresh ChatGPT conversation, attach or paste the useful history, and continue the same work there.
The Export & new chat button downloads the Markdown export and opens a new ChatGPT tab.
Markdown exports have three detail levels:
Clean keeps the user tasks and final visible responses.
Progress, the default, also keeps useful assistant progress and task plans while leaving out raw tool-call noise.
Full keeps the detailed records and tool calls for cases where exact commands or payloads are useful.
Conversation processing, archive storage, counters, history rendering, and Markdown export happen locally in the browser. AntiCurse does not send your conversation to the developer or to another service, and it does not modify the server-side ChatGPT conversation.
GPT AntiCurse is an independent project and is not affiliated with or endorsed by OpenAI.
Large coding, research, debugging, and agentic-work chats can eventually contain hundreds or thousands of turns plus tool calls, hidden records, branches, and other internal conversation state. Even when the model is idle, processing all of that state can make the ChatGPT tab noticeably slower.
AntiCurse keeps a smaller recent part of the conversation active in ChatGPT instead of making the frontend process the complete historical graph every time.
There are two history modes.
Recent N + button keeps a configurable recent window in ChatGPT's normal interface. When you need older messages, a Load previous button appears above the conversation.
Auto window uses the same bounded recent history, but loads older archived messages automatically as you scroll toward the beginning of the chat.
Older history is kept outside ChatGPT's main React conversation state and rendered using a lightweight, virtualized view. Only a small number of archived pages stay mounted at once, so loading more history does not simply recreate the original performance problem.
AntiCurse can also keep a local backup of the conversation and export it to Markdown.
This is particularly useful for long-running projects. If a conversation eventually becomes too large for the model's useful context, you can export it, open a fresh ChatGPT conversation, attach or paste the useful history, and continue the same work there.
The Export & new chat button downloads the Markdown export and opens a new ChatGPT tab.
Markdown exports have three detail levels:
Clean keeps the user tasks and final visible responses.
Progress, the default, also keeps useful assistant progress and task plans while leaving out raw tool-call noise.
Full keeps the detailed records and tool calls for cases where exact commands or payloads are useful.
Conversation processing, archive storage, counters, history rendering, and Markdown export happen locally in the browser. AntiCurse does not send your conversation to the developer or to another service, and it does not modify the server-side ChatGPT conversation.
GPT AntiCurse is an independent project and is not affiliated with or endorsed by OpenAI.
评分 0（1 位用户）
权限与数据
更多信息