ck-pi-cline0.1.11

Cline Account API provider, official client header disguise, free model catalog synchronization, reverse proxy, and session management for the Pi coding agent.

cknb666 Published 5 days ago MIT ~800 downloads/mo

Untested on omp

This is a pi package. No one has reported whether it works on omp yet. If you try it, add your result below.

Reports 0

No reports yet.

Readme

ck-pi-cline

中文说明 | English Documentation

赞助与支持

如果您觉得本项目提升了您的开发效率，欢迎赞助支持！您的赞助将直接用于持续维护、网络节点验证、逆向协议对齐与新特性的快速迭代。

赞助渠道

GitHub Sponsors: https://github.com/sponsors/1837620622
企业赞助与商务合作:
- 欢迎 AI 基础设施提供商、代理网关服务商或团队提供赞助。
- 赞助权益包含：在项目 README 显著位置展示企业 Logo 与链接、优先处理定制化需求与专属技术支持。

一、项目概述

ck-pi-cline 是专为 Pi Coding Agent 打造的高可用 Cline 官方 API 逆向代理、客户端指纹伪装、全量零额度模型智能路由与本地 OpenAI 兼容反向代理插件。

架构与运行原则：

默认原生极简直连与模型注入：启动运行时默认不开启本地反代端口，直接在 Pi 运行时中以 Cline (Free) 供应商注入 17 款零额度免费模型，通过全局 Fetch 拦截器直连官方网关（零本地代理开销，架构与 ck-pi-zen-session 完全对齐）；

按需启动本地反向代理：仅在显式执行 /cline proxy start 命令或需要供第三方工具（如 CC-Switch、Cursor）使用时才启动本地 4116 端口，可通过 /cline proxy stop 随时停止并恢复直连；

同模型指数退避重试：遭遇 500/502/503/504/429 报错时，坚守模型质量底线，自动进行同模型指数退避重试，绝不擅自降级或切换低配备用模型。

在实际使用官方 Cline 接口时，开发者普遍面临以下技术痛点：

防盗刷指纹拦截：非官方客户端直接请求 https://api.cline.bot/api/v1 会遭遇 403 阻断或限制；
响应格式不兼容：Cline 官方接口在非流式模式下将结果包裹于 { success: true, data: { choices: [...] } }，直接破坏了原生 OpenAI SDK 与各类 IDE 适配器的预期；
节点异常引发客户端崩溃：上游推理集群满载或故障时，在 text/event-stream 中返回 choices: [] 且下发内部错误帧，触发 Pi 等智能体运行时致命错误 model output must contain either output text or tool calls, these cannot both be empty；
同名模型扣费陷阱：例如历史模型若遗漏 :free 标识，上游将走商业计费通道扣除账户余额；
模型下架与变动频繁：部分旧模型（如 deepseek-v4.1-flash, space-bunny-alpha, ling-3.0-flash-fin 等）已被官方下架或转为付费。

ck-pi-cline 从协议层、拦截层与代理层彻底解决了上述所有问题，实现即插即用、完全零额度消耗与高可靠容灾保障。

二、系统架构与技术原理解析

+-----------------------------------------------------------------------------------+
|                            Client Integration Layer                               |
|   Pi Terminal Agent  |  CC-Switch  |  Cursor IDE  |  Cherry Studio  |  OpenAI SDK |
+-----------------------------------------------------------------------------------+
                                         | (HTTP / SSE Requests)
                                         v
+-----------------------------------------------------------------------------------+
|                 ck-pi-cline Hook & Local Proxy Layer                              |
|                                                                                   |
|  [1. Fingerprint Injector]                                                        |
|      - x-client-version: 4.1.16 | User-Agent: Cline/4.1.16 | vscode headers       |
|                                                                                   |
|  [2. Model Alias & Zero-Cost Guard]                                               |
|      - Short Aliases (free, ling, 550b, lightning, laguna, etc.)                  |
|      - Auto append :free protection to shield credits                             |
|      - Whitelist 17 verified zero-cost active models                              |
|                                                                                   |
|  [3. Context Pruning & Audit Sanitization]                                        |
|      - Truncate giant tool logs (> 25,000 chars)                                  |
|      - Sanitize SQLi/XSS/exploit payloads in <conversation> against content_filter|
|                                                                                   |
|  [4. SSE Stream Transformer & Empty Output Guard]                                 |
|      - Intercept choices: [] & error frames                                       |
|      - Normalize delta.reasoning -> delta.reasoning_content                       |
|      - Force synthetic chunk on stream close if empty                             |
|                                                                                   |
|  [5. Same-Model Exponential Backoff Retry]                                        |
|      - Transparent retry on 500/502/503/504/429 preserving target model & context|
+-----------------------------------------------------------------------------------+
                                         | (Disguised HTTPS Upstream)
                                         v
+-----------------------------------------------------------------------------------+
|                        Cline Cloud Gateway (api.cline.bot)                        |
|                                         |
|       +---------------------------------+---------------------------------+
|       v                                 v                                 v
| [OpenRouter Infrastructure]   [Novita AI GPU Cluster]   [NVIDIA NIM Distributed]  |
+-----------------------------------------------------------------------------------+

1. 官方客户端特征标头伪装 (Fingerprint Spoofing)

Cline 网关部署了反作弊指纹校验。插件在所有出站请求中精准伪造官方 VS Code 扩展的完整指纹特征：

User-Agent: Cline/4.1.16
x-client-version: 4.1.16
x-core-version: 4.1.16
x-platform-version: 1.106.0
x-client-type: cline-vscode
http-referer: https://cline.bot
x-platform: vscode
x-title: Cline

2. 响应体解包与协议归一化 (Protocol Normalization)

Cline 官方非流式端点会将 OpenAI 规范的 Completion 对象嵌套在 data 属性内。标准客户端找不到根节点的 choices 会抛出异常。插件在反向代理与 Fetch 拦截层自动执行透明解包：

提取 rawJson.data 提升至根节点；
完整保留 id, model, choices, usage 等标准字段；
对外暴露符合标准 RFC 的 OpenAI ChatCompletion 数据格式。

3. 空输出死锁防御机制 (Empty Output Guard Invariant)

在上游节点高载荷时，网关偶发返回空流，触发 Pi 的致命断言失败。插件通过管道分析 SSE 帧：若遭遇异常错误帧或流结束时无文本也无工具调用，自动注入保底响应帧，确保调用链不崩溃并友好提示用户重试。

4. 隐身模型识别与零额度白嫖判定机理 (Stealth Models & Zero-Cost Guard)

自动识别官方零扣费模型（如 openrouter/free），对同名模型自动追加 :free 防护后缀，拦截历史下架付费陷阱。

5. 深度思考推理链映射 (Reasoning Tokens Alignment)

自动双向归一化 delta.reasoning 与 delta.reasoning_content，保障思考折叠块在各类前端与终端正常渲染。

6. 同模型指数退避重试机制 (Same-Model Exponential Backoff Retry)

遭遇 500、502、503、504、524 或 429 限流时，坚守模型质量底线，自动进行最多 3 轮指数退避重试，绝不擅自降级或切换低配备用模型。

7. 全链路上下文超限修剪 (Context Pruning Engine)

拦截单个超过 25,000 字符的巨型终端输出，修剪 <conversation> 与 <previous-summary> 中的冗余中间历史，避免触发 400 Context Length Exceeded。

8. 安全审核过滤阻断防御与 Token Cap 修复 (content_filter & Token Cap Defense)

审计日志脱敏：引入 sanitizeSensitiveAuditContent，对压缩历史中的 SQL 注入、XSS 脚本块、漏洞载荷进行占位脱敏，从源头消除上游内容审查拦截；
finish_reason content_filter 平滑恢复：捕获内容审核过滤帧，在压缩模式下合成安全检查点摘要，在对话模式下保全已生成文本并附带合规指引；
Token Cap 自动续接：将压缩请求上限提升至 16,384 tokens 并压制 CoT 思考耗时；当命中 finish_reason: "length" 时自动改写为 "stop"，彻底解决 Summarization failed: generation hit the token cap 异常。

三、实测 17 款零额度模型全量矩阵

所有收录模型均经过真实在线请求验证，确认额度消耗为 0：

1. 隐身与智能路由零消耗模型 (Stealth & Smart Routers)

模型 ID	简写别名	上下文窗口	最大输出	思考推理	特性与适用场景
`openrouter/free`	`free`, `auto`	200,000	32,768	支持	OpenRouter 官方高可用免费聚合节点，首选兜底

2. 百万级超长上下文免费模型 (1M+ Giant Context)

模型 ID	简写别名	上下文窗口	最大输出	思考推理	特性与适用场景
`nvidia/nemotron-3-ultra-550b-a55b:free`	`550b`, `ultra`	1,000,000 (1M)	32,768	支持	550B 参数超大规模模型，通用推理与知识检索
`nvidia/nemotron-3.5-lightning:free`	`lightning`	1,000,000 (1M)	32,768	支持	1M 上下文轻量化极速变体
`thinkingmachines/inkling:free`	`inkling`	1,048,576 (1M)	32,768	支持	1M 上下文长文本分析模型
`thinkingmachines/inkling-small:free`	`inkling-small`	1,048,576 (1M)	32,768	支持	1M 轻量长文本模型

3. 主力代码与深度思考免费模型 (Core Coding & Reasoning)

模型 ID	简写别名	上下文窗口	最大输出	思考推理	特性与适用场景
`inclusionai/ling-3.0-flash-sante:free`	`ling`, `sante`	262,144	32,768	支持	极速响应，稳定性高的高可用主力模型
`nvidia/nemotron-3-super-120b-a12b:free`	`120b`	262,144	32,768	支持	120B 参数级深度思考模型
`nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free`	`nano`, `reasoning`	131,072	32,768	支持 (带视觉)	多模态图像识别 + 深度思考链路
`cohere/north-mini-code:free`	`north`	131,072	32,768	支持	代码补全、单元测试与代码审查专用
`poolside/laguna-s-2.1:free`	`laguna`	131,072	32,768	支持	算法与逻辑推导优化模型
`poolside/laguna-xs-2.1:free`	`xs`	131,072	32,768	支持	轻量级逻辑变体

4. 轻量极速与专用工具模型 (Lightweight & Specialized Tools)

模型 ID	简写别名	上下文窗口	最大输出	思考推理	特性与适用场景
`google/gemma-4-26b-a4b-it:free`	`gemma`	262,144	32,768	支持	谷歌 Gemma 开源免费微调版
`google/gemma-4-31b-it:free`	`gemma-31b`	262,144	32,768	支持	谷歌 Gemma 31B 结构化推理版
`dots-studio/dots-3-note-preview:free`	`note`	131,072	32,768	支持	笔记、提纲与纪要总结优化
`liquid/lfm-2.5-2.6b:free`	`lfm`	32,768	8,192	快速	Liquid 神经架构轻量模型
`nvidia/nemotron-3.5-content-safety:free`	`safety`	131,072	16,384	快速	内容安全审计过滤专用
`apodex/apodex-1.1-mini:free`	`apodex`	131,072	32,768	快速	极速辅助对话模型

5. 计费陷阱与已下架模型警示 (Paid Traps & Delisted Models)

以下模型已被上游下架或转为商业计费，插件已将其列入防御过滤，自动重定向至可用模型：

deepseek/deepseek-v4.1-flash: 官方已停止免费分发，已自动映射至 openrouter/free；
stealth/space-bunny-alpha: 上游节点已下架，已自动平滑重定向；
openrouter/fusion / openrouter/pareto-code: 节点已停用，已自动平滑重定向；
inclusionai/ling-3.0-flash-fin:free: 历史版本已下架，已自动升级至 inclusionai/ling-3.0-flash-sante:free；
qwen/qwen3.8-27b:free: 上游免费端点不可用，已自动重定向至 openrouter/free。

四、命令使用与交互面板

在 Pi 终端交互界面中，可通过 /cline 体系指令进行全方位控制：

命令	说明	交互效果
`/cline`	无参直接运行：自动在线探测拉取后端最新模型，刷新上下文与思考等级，同步配置并热载入 Pi	输出控制面板与已同步模型摘要
`/cline <key>` 或 `/cline key <key>`	更新 API Key：保存新 Key，自动在线探测拉取最新免费模型并热重载	输出新 Key 验证结果与模型卡片
`/cline refresh` 或 `/cline sync`	强制在线刷新：重新在线拉取远端模型目录并全量同步本地配置与运行时	输出同步状态报告
`/cline list` 或 `/cline models`	查看模型列表：展示全部 17 款零额度模型详细树形卡片	输出分级卡片列表
`/cline free`	查看零额度注册表：分类呈现隐身零消耗模型与官方标准免费模型	输出分类结构表格
`/cline status`	查看运行状态：脱敏 Key、端点、可用模型总数、思考模型数及本地反代状态	输出运行状态仪表盘
`/cline ping [模型ID]`	网络时延探针：实时探测免费模型连通性与往返延迟 (RTT)	输出各模型网络时延与评级
`/cline model <模型ID>`	快速切换模型：一键切换当前 Pi 会话所使用的模型	即刻切换生效
`/cline proxy start [端口]`	启动本地反向代理：在指定端口（默认 4116）启动 OpenAI 兼容本地代理	输出端点信息
`/cline proxy stop`	停止本地反向代理：停止正在运行的代理进程	释放端口

五、本地反向代理配置与第三方集成

插件内置 Node.js 反向代理服务，将 Cline 官方私有接口转换为完全符合标准 OpenAI 规范的本地服务。

1. 独立 CLI 启动代理

可在终端直接执行打包后的可执行文件：

# 默认在 4116 端口启动
node ./bin/cline-proxy.js 4116

# 或通过自定义参数启动
node ./bin/cline-proxy.js --port 4116 --key sk_test_placeholder_key

2. CC-Switch 客户端接入

在 CC-Switch 中添加自定义 OpenAI 供应商：

Provider Name: Cline-Free
Base URL: http://127.0.0.1:4116/v1
API Key: 填入任意字符串（如 cline-proxy）
Models: 可通过 /v1/models 端点一键自动发现。

3. Cursor / VS Code 接入

在 Cursor 设置的 OpenAI API Key 中：

Override OpenAI Base URL: http://127.0.0.1:4116/v1
API Key: 任意填写
Model Name: 手动输入 openrouter/free 或 nvidia/nemotron-3.5-lightning:free。

4. OpenAI SDK (Python / Node.js) 接入

Python:

from openai import OpenAI

client = OpenAI(
    base_url="http://127.0.0.1:4116/v1",
    api_key="cline-local-token"
)

response = client.chat.completions.create(
    model="openrouter/free",
    messages=[{"role": "user", "content": "写一段关于分布式锁的最佳实践"}],
    stream=True
)

for chunk in response:
    content = chunk.choices[0].delta.content
    if content:
        print(content, end="", flush=True)

Node.js:

import OpenAI from "openai";

const openai = new OpenAI({
  baseURL: "http://127.0.0.1:4116/v1",
  apiKey: "cline-local-token",
});

const completion = await openai.chat.completions.create({
  model: "openrouter/free",
  messages: [{ role: "user", content: "Implement a Red-Black Tree in TypeScript" }],
});

console.log(completion.choices[0].message.content);

六、版本变更记录 (Changelog)

v0.1.9

全新免费模型深度对接与别名路由强化 (New Free Models & Alias Enhancement)：
- 正式收录并适配最新 apodex/apodex-1.1-mini:free 等活跃免费模型，强化 dots-studio/dots-3-note-preview:free、cohere/north-mini-code:free 等代码与推理模型；
- 针对上游已取消免费标识的 inclusionai/ling-3.0-flash-sante，路由层平滑安全重定向至 openrouter/free，彻底杜绝扣费与 404 隐患；
- 增强短别名支持（dots, apodex, laguna, north, gemma 等）。
会话压缩深度自愈与 Token Cap 彻底防御 (Auto-compaction Token Cap Boost)：
- 针对 Pi 压缩模块传入的极小 max_tokens（通常仅 1000~2000 tokens），拦截层与本地反向代理双重强制提升至至少 16384 tokens，从源头根除 Summarization failed: generation hit the token cap 异常；
- 强化流式与非流式 finish_reason: "content_filter" 与 finish_reason: "length" 的捕获，平滑重写为 stop 并兜底输出合规的结构化会话进度检查点，确保压缩任务 100% 成功。

v0.1.8

全链路会话压缩与安全策略防御 (content_filter & Token Cap Fix)：
- 修复 Token Cap 截断异常：调优压缩任务输出上限至 16,384 tokens，并在压缩模式下强制设置 reasoning_effort: "low" 并剥离思考过程，彻底消除推理模型触发 token cap 导致压缩失败；
- 平滑重写 finish_reason length -> stop：在压缩摘要非空时将 length 结果重写为 stop，避免 Pi 触发 stopReason === "length" 运行时报错；
- 安全审核过滤防御：集成 sanitizeSensitiveAuditContent 审计脱敏引擎，自动合规替换 <conversation> 中的 SQL 注入、XSS 脚本标签与利用工具特征；
- 拦截器与本地代理双向 content_filter 兜底：在压缩时合成结构化会话检查点，在对话时保全已有输出并提示调整提问方式。
免费模型注册表更新与别名路由升级：
- 清除上游已失效或下架的 6 款模型，精准锁定 17 款在线零扣费模型；
- 自动将旧模型别名（如 bunny, fusion, code, deepseek-v4.1-flash, ling）平滑重定向至当前活跃免费模型。

v0.1.7

安全审计与网络加固：
- 增强本地代理 Host 标头安全校验与 DNS 重绑定防护；
- 统一请求体重试与多字节 UTF-8 流分包保障。

v0.1.6

深度修复上游空响应重试机制 (Provider Returned an Empty Response)：
- 流式预读上限提升至 30 包与 64KB，精准穿透网关 processing 注释；
- 严格限定错误边界匹配，生产环境全抖动指数退避重试。

v0.1.5

供应商节点空响应自动同模型重试 3 次：首创首包无损流式 Stream Peek 预读探测，当上游节点返回伪 200 空流时自动 3 次重试；
Pi 运行态规范化校验 (sanitizePiModelDefinition)：严格清洗畸变字段，保证配置安全落盘。

v0.1.4

系统安全审计全面加固：DNS Rebinding 防护、受信任 CORS 来源校验、0700/0600 权限保障。

v0.1.2

坚守模型品质底线，严格执行同模型指数退避重试，绝不自动跨模型降级；极简默认免代理直连模式。

v0.1.1

指令体系与 /zen 全面深度对齐；思考链 (CoT) 双向字段规范化；清除所有 Emoji 表情符号。

v0.1.0

全新发布！Cline 官方指纹伪装、全量零额度免费模型智能路由与本地 OpenAI 兼容反向代理。

七、许可

仓库: https://github.com/1837620622/ck-pi-extension
协议: MIT License

ck-pi-cline (English)

中文说明 | English Documentation

Sponsorship & Support

If you find this project helpful, please consider sponsoring. Your support directly funds continuous maintenance, endpoint testing, reverse-engineering protocol alignment, and rapid iteration.

Sponsorship Channels

GitHub Sponsors: https://github.com/sponsors/1837620622
Enterprise Sponsorship:
- AI infrastructure providers, proxy gateway services, and engineering teams are welcome to sponsor.
- Benefits: Prominent brand logo and link display in the README, priority issue response, and custom protocol optimizations.

1. Overview
2. System Architecture & Technical Principles
3. Verified 17 Zero-Cost Models Matrix
4. Commands & Dashboard
5. Local Reverse Proxy & Third-Party Integration
6. Changelog
7. License

1. Overview

ck-pi-cline is a high-availability Cline official API reverse proxy, client fingerprint disguise, zero-cost model intelligent router, and local OpenAI-compatible reverse proxy extension built for the Pi Coding Agent.

Architecture & Operational Principles:

Default Direct Connection & Injection: By default, the local proxy port is not started upon launch. The extension directly mounts 17 zero-cost models into Pi runtime via a global Fetch interceptor pointing to the official gateway with zero local proxy overhead (fully aligned with ck-pi-zen-session);

On-Demand Local Reverse Proxy: Starts the local port 4116 only when explicitly executed via /cline proxy start or when needed by external tools (e.g., CC-Switch, Cursor);

Same-Model Exponential Backoff Retry: Transparently retries errors (500/502/503/504/429) on the exact same model, refusing to downgrade model capability.

Key problems resolved:

Fingerprint Validation Blocking: Bypasses 403 errors by spoofing official VS Code headers;
Payload Discrepancies: Unpacks non-streaming responses nested in { data: { choices: [...] } };
Empty Output Freezes: Intercepts choices: [] frames to prevent agent crashes;
Credit Depletion Traps: Excludes paid models and enforces :free parameters;
Model Deprecations: Cleans delisted models and dynamically redirects aliases to working free alternatives.

2. System Architecture & Technical Principles

Official Client Fingerprint Spoofing: Injects 8 official headers (User-Agent: Cline/4.1.16, x-client-version: 4.1.16, x-client-type: cline-vscode, etc.) across all outgoing requests.
Protocol Normalization: Transparently unpacks { success: true, data: { ... } } into standard OpenAI ChatCompletion objects.
Empty Output Invariant Guard: Inspects SSE frames and synthesizes fallback assistant chunks when the upstream returns empty choices.
Zero-Cost Model Safeguard: Detects verified free models and appends :free to prevent billable routing.
Reasoning Chain Mapping: Normalizes delta.reasoning into delta.reasoning_content for proper CoT rendering.
Same-Model Exponential Backoff Retry: Retries up to 3 times on 5xx and 429 gateway responses without model substitution.
Context Pruning: Truncates oversized tool outputs (> 25,000 chars) and trims intermediate conversation turns.
content_filter Defense & Token Cap Fix:
- Sanitizes SQLi, XSS, and security payloads in <conversation> to prevent triggering safety moderation;
- Intercepts finish_reason: "content_filter" and normalizes to structured checkpoint summaries;
- Relaxes compaction token limit to 16,384 tokens with reasoning_effort: "low", rewriting finish_reason: "length" to "stop" to eliminate summarization crashes.

3. Verified 17 Zero-Cost Models Matrix

All listed models have been tested online with verified zero credit deduction (cost === 0):

Model ID	Short Alias	Context Window	Max Output	Reasoning	Modality
`openrouter/free`	`free`, `auto`	200,000	32,768	Yes	Text
`nvidia/nemotron-3-ultra-550b-a55b:free`	`550b`, `ultra`	1,000,000 (1M)	32,768	Yes	Text
`nvidia/nemotron-3.5-lightning:free`	`lightning`	1,000,000 (1M)	32,768	Yes	Text
`thinkingmachines/inkling:free`	`inkling`	1,048,576 (1M)	32,768	Yes	Text
`thinkingmachines/inkling-small:free`	`inkling-small`	1,048,576 (1M)	32,768	Yes	Text
`inclusionai/ling-3.0-flash-sante:free`	`ling`, `sante`	262,144	32,768	Yes	Text
`nvidia/nemotron-3-super-120b-a12b:free`	`120b`	262,144	32,768	Yes	Text
`nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free`	`nano`, `reasoning`	131,072	32,768	Yes	Text, Vision
`cohere/north-mini-code:free`	`north`	131,072	32,768	Yes	Text
`poolside/laguna-s-2.1:free`	`laguna`	131,072	32,768	Yes	Text
`poolside/laguna-xs-2.1:free`	`xs`	131,072	32,768	Yes	Text
`google/gemma-4-26b-a4b-it:free`	`gemma`	262,144	32,768	Yes	Text
`google/gemma-4-31b-it:free`	`gemma-31b`	262,144	32,768	Yes	Text
`dots-studio/dots-3-note-preview:free`	`note`	131,072	32,768	Yes	Text
`liquid/lfm-2.5-2.6b:free`	`lfm`	32,768	8,192	Fast	Text
`nvidia/nemotron-3.5-content-safety:free`	`safety`	131,072	16,384	Fast	Text
`apodex/apodex-1.1-mini:free`	`apodex`	131,072	32,768	Fast	Text

4. Commands & Dashboard

Control extension features inside the Pi interactive terminal:

Command	Description	Output
`/cline`	No args: discover remote models, sync configuration, and hot reload Pi	Status panel and synced summary
`/cline <key>` or `/cline key <key>`	Configure API key and reload models	Key verification report
`/cline refresh` or `/cline sync`	Force remote sync and update local configuration	Refresh status report
`/cline list` or `/cline models`	View catalog of 17 zero-cost models	Tree card list
`/cline free`	View categorized zero-cost models table	Structured catalog table
`/cline status`	View API key mask, endpoint, model count, and proxy state	Operational dashboard
`/cline ping [model-id]`	Probe free model network connectivity and round-trip time	Latency and rating report
`/cline model <model-id>`	Switch active Cline model	Immediate activation
`/cline proxy start [port]`	Start local OpenAI-compatible proxy (default port 4116)	Endpoint details
`/cline proxy stop`	Stop local proxy service	Port release

5. Local Reverse Proxy & Third-Party Integration

Standalone CLI Proxy

node ./bin/cline-proxy.js 4116

CC-Switch

Provider: Cline-Free
Base URL: http://127.0.0.1:4116/v1
API Key: cline-proxy

OpenAI SDK Integration

from openai import OpenAI

client = OpenAI(
    base_url="http://127.0.0.1:4116/v1",
    api_key="cline-local-token"
)

response = client.chat.completions.create(
    model="openrouter/free",
    messages=[{"role": "user", "content": "Hello!"}],
    stream=True
)
for chunk in response:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)

6. Changelog

v0.1.9

New Free Models Integration & Enhanced Routing:
- Registered and calibrated newly active free models such as apodex/apodex-1.1-mini:free, dots-studio/dots-3-note-preview:free, and cohere/north-mini-code:free;
- Safely and smoothly routed unflagged inclusionai/ling-3.0-flash-sante to openrouter/free to eliminate unexpected charges or 404s;
- Expanded shorthand aliases (dots, apodex, laguna, north, gemma, etc.).
Auto-compaction Token Cap Boost & Crash Defense:
- Compaction requests dynamically boost max_tokens / max_completion_tokens to at least 16,384 tokens with reasoning_effort: "low" to eliminate token cap exhaustion crashes;
- Both interceptor and proxy intercept and normalize finish_reason: "content_filter" and finish_reason: "length" to "stop", synthesizing structured checkpoints when needed.

v0.1.8

Auto-compaction & content_filter defense:
- Relaxed compaction token output limit to 16,384 tokens with reasoning_effort: "low" and CoT suppression, preventing token cap exhaustion crashes;
- Normalized finish_reason: "length" on compaction tasks to "stop";
- Added sanitizeSensitiveAuditContent to sanitize SQL injection, XSS script tags, and exploit patterns in <conversation> tags;
- Intercepted finish_reason: "content_filter", synthesizing structured session checkpoints on compaction.
Model catalog & alias routing update:
- Purged 6 delisted/paid trap models, locking catalog to 17 verified active models;
- Smooth alias auto-routing for deprecated models.

v0.1.7

Strengthened Host header validation, DNS rebinding defenses, and cross-chunk UTF-8 stream decoding.

v0.1.6

Deep fix for upstream empty response retry (Provider returned an empty response) with 30-packet peek buffer.

v0.1.5

3-attempt same-model retry on empty response or zero-byte stream; sanitizePiModelDefinition data protection.

v0.1.4

Comprehensive security audit fixes: DNS rebinding protection, trusted CORS origin validation, 0700/0600 file permissions.

v0.1.2

Strict same-model exponential backoff retry; default direct connection mode.

v0.1.1

Command parity with /zen; bidirectional CoT reasoning mapping; stripped all emojis.

v0.1.0

Initial release with official fingerprint spoofing and verified zero-cost models.

7. License

Repository: https://github.com/1837620622/ck-pi-extension
License: MIT License

Untested on omp

Did it work for you?

Reports 0

Readme

ck-pi-cline

赞助与支持

赞助渠道

目录

一、项目概述

二、系统架构与技术原理解析

1. 官方客户端特征标头伪装 (Fingerprint Spoofing)

2. 响应体解包与协议归一化 (Protocol Normalization)

3. 空输出死锁防御机制 (Empty Output Guard Invariant)

4. 隐身模型识别与零额度白嫖判定机理 (Stealth Models & Zero-Cost Guard)

5. 深度思考推理链映射 (Reasoning Tokens Alignment)

6. 同模型指数退避重试机制 (Same-Model Exponential Backoff Retry)

7. 全链路上下文超限修剪 (Context Pruning Engine)

8. 安全审核过滤阻断防御与 Token Cap 修复 (content_filter & Token Cap Defense)

三、实测 17 款零额度模型全量矩阵

1. 隐身与智能路由零消耗模型 (Stealth & Smart Routers)

2. 百万级超长上下文免费模型 (1M+ Giant Context)

3. 主力代码与深度思考免费模型 (Core Coding & Reasoning)

4. 轻量极速与专用工具模型 (Lightweight & Specialized Tools)

5. 计费陷阱与已下架模型警示 (Paid Traps & Delisted Models)

四、命令使用与交互面板

五、本地反向代理配置与第三方集成

1. 独立 CLI 启动代理

2. CC-Switch 客户端接入

3. Cursor / VS Code 接入

4. OpenAI SDK (Python / Node.js) 接入

六、版本变更记录 (Changelog)

v0.1.9

v0.1.8

v0.1.7

v0.1.6

v0.1.5

v0.1.4

v0.1.2

v0.1.1

v0.1.0

七、许可

ck-pi-cline (English)

Sponsorship & Support

Sponsorship Channels

Table of Contents

1. Overview

2. System Architecture & Technical Principles

3. Verified 17 Zero-Cost Models Matrix

4. Commands & Dashboard

5. Local Reverse Proxy & Third-Party Integration

Standalone CLI Proxy

CC-Switch

OpenAI SDK Integration

6. Changelog

v0.1.9

v0.1.8

v0.1.7

v0.1.6

v0.1.5

v0.1.4

v0.1.2

v0.1.1

v0.1.0

7. License