智能助手网
标签聚合 ccswitch

/tag/ccswitch

linux.do · 2026-04-16 13:19:54+08:00 · tech

纯小白写个白嫖的codex账号怎么用于cc的流程,这两天在L站搜了很久,好像没有系统性介绍的,自己摸索了下,没啥技术含量,资深佬友可以略过: 1、整个路径流程是这样的:L站各种注册机 CPA反代 ccswitch​ vscode​ claudecode; 2、我原本以为CPA里面的是OpenAI的格式要转换成Anthropic支持的格式,站里也有各位佬友的教程,其实不需要,ccswitch就能实现; 3、大概流程截图下: 1 个帖子 - 1 位参与者 阅读完整话题

linux.do · 2026-04-14 00:35:58+08:00 · tech

我本人在公司开发基本上不用第三方,自己买了很多号自己切。我不喜欢用CPA反带出来,就喜欢直接登录官方。在CC switch里可以轻易替换,但时间久了会过期,所以说我就做了一个CPA换到CCS switch的一个转换工具 我发现在CC switch里切号只需要把token什么信息填进去,就可以直接切不同的官方号之前一直是手动复制,感觉有点傻,而且token很长很费我的鼠标手。我就写了一个转换工具 地址如下 http://loveyless.de5.net:50515/ 嗯,我是自己部署的,大家可以用提示词快速写一个,不过这个没什么用场景太单一了 提示词 (点击了解更多详细信息) 6 个帖子 - 4 位参与者 阅读完整话题

linux.do · 2026-04-13 22:00:09+08:00 · tech

我现在在ccswitch的配置 model_provider = “custom” model = “gpt-5.4” service_tier = “fast” network_access = “enabled” windows_wsl_setup_acknowledged = true model_reasoning_effort = “xhigh” disable_response_storage = true model_context_window = 1000000 model_auto_compact_token_limit = 900000 [model_providers] [model_providers.custom] name = “custom” wire_api = “responses” requires_openai_auth = true base_url = “ https://ai.spring.top ” 通过接口直接调用是可以的 sub2api里测试也是可以的 2 个帖子 - 2 位参与者 阅读完整话题

linux.do · 2026-04-13 16:29:12+08:00 · tech

在Claude code中,如果直接利用ccswitch使用DeepSeek-V3.2,会经常碰到如下报错: API Error: 400 {"error":{"message":"This model's maximum context length is 102400 tokens. However, you requested 102945 tokens (70945 in the messages, 32000 in the completion). Please reduce the length of the messages or completion.","type":"invalid_request_error","param":null,"code":"invalid_request_error"}} 原因是Claude code映射的模型支持的上下文都远远超过DeepSeek-V3.2的128k,解决方案是设置"CLAUDE_CODE_AUTO_COMPACT_WINDOW"参数,注意此参数需要设置在env块中,否则无效。 示例: { "env": { "ANTHROPIC_AUTH_TOKEN": "sk-xxxxxxx", "ANTHROPIC_BASE_URL": "https://api.deepseek.com/anthropic", "ANTHROPIC_DEFAULT_HAIKU_MODEL": "DeepSeek-V3.2", "ANTHROPIC_DEFAULT_OPUS_MODEL": "DeepSeek-V3.2", "ANTHROPIC_DEFAULT_SONNET_MODEL": "DeepSeek-V3.2", "ANTHROPIC_MODEL": "DeepSeek-V3.2", "CLAUDE_CODE_MAX_OUTPUT_TOKENS": 63900, "CLAUDE_CODE_AUTO_COMPACT_WINDOW": 128000, }, "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": 1, "language": "中文" } 另外,"CLAUDE_AUTOCOMPACT_PCT_OVERRIDE"参数目前实测无效,Autocompact buffer固定在33k,官方的仓库中的issue也对此有所讨论: github.com/anthropics/claude-code CLAUDE_AUTOCOMPACT_PCT_OVERRIDE not triggering at configured threshold 已打开 10:12PM - 19 Mar 26 UTC 已关闭 08:59PM - 20 Mar 26 UTC YebooLixoo bug has repro platform:macos area:core ## Bug Description `CLAUDE_AUTOCOMPACT_PCT_OVERRIDE=55` is set as an environmen … t variable but auto-compaction does not trigger at 55% context usage. The main session regularly exceeds the configured threshold without compacting — observed at 67% in one session and over 80% in another before compaction finally occurred. ## Environment - **Claude Code version**: 2.1.79 - **Model**: Claude Opus 4.6 (Max plan) - **OS**: macOS (Darwin 23.5.0) - **Shell**: zsh ## Steps to Reproduce 1. Set `export CLAUDE_AUTOCOMPACT_PCT_OVERRIDE=55` in `~/.zshrc` 2. Verify the variable is loaded: `echo $CLAUDE_AUTOCOMPACT_PCT_OVERRIDE` → `55` 3. Start a Claude Code session and use it until context exceeds 55% 4. Observe that auto-compaction does **not** trigger at 55% ## Expected Behavior Auto-compaction should trigger when context window usage reaches ~55%, as configured by `CLAUDE_AUTOCOMPACT_PCT_OVERRIDE`. ## Actual Behavior - Context reached **67%** (see screenshot) without triggering auto-compaction - In another session, context reached **over 80%** before auto-compaction finally fired - The env var appears to be ignored for the main session ## Screenshot Context at 67% with no auto-compaction triggered: <img width="816" height="198" alt="Image" src="https://github.com/user-attachments/assets/2cd4a39f-3b6d-4144-bd69-43a82fc843f1" /> ## Additional Context - The env var is confirmed set in the shell environment - `CLAUDE_CODE_AUTO_COMPACT_WINDOW` is not set (default) - According to the docs, `CLAUDE_AUTOCOMPACT_PCT_OVERRIDE` should apply to both main conversations and subagents - The HUD percentage comes from Claude Code's native `used_percentage` field, so the reported percentage should match what the compaction engine sees 1 个帖子 - 1 位参与者 阅读完整话题