openclaude-multi-llm

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

OpenClaude Multi-LLM Skill

OpenClaude 多LLM Skill

Skill by ara.so — Daily 2026 Skills collection.

OpenClaude is a fork of Claude Code that routes all LLM calls through an OpenAI-compatible shim (

openaiShim.ts

), letting you use any model that speaks the OpenAI Chat Completions API — GPT-4o, DeepSeek, Gemini via OpenRouter, Ollama, Groq, Mistral, Azure, and more — while keeping every Claude Code tool intact (Bash, FileRead, FileWrite, FileEdit, Glob, Grep, WebFetch, Agent, MCP, Tasks, LSP, NotebookEdit).

由 ara.so 开发的Skill — 2026年度每日Skill合集。

OpenClaude是Claude Code的分支版本，它通过一个兼容OpenAI协议的垫片（

openaiShim.ts

）路由所有LLM调用，让你可以使用任何支持OpenAI聊天补全API的模型——包括GPT-4o、DeepSeek、通过OpenRouter接入的Gemini、Ollama、Groq、Mistral、Azure等——同时完整保留Claude Code的所有工具能力（Bash、FileRead、FileWrite、FileEdit、Glob、Grep、WebFetch、Agent、MCP、Tasks、LSP、NotebookEdit）。

Installation

安装

npm (recommended)

npm（推荐方式）

bash

npm install -g @gitlawb/openclaude

bash

npm install -g @gitlawb/openclaude

CLI command installed: openclaude

安装后的CLI命令：openclaude

undefined

undefined

From source (requires Bun)

源码安装（需要Bun环境）

bash

git clone https://node.gitlawb.com/z6MkqDnb7Siv3Cwj7pGJq4T5EsUisECqR8KpnDLwcaZq5TPr/openclaude.git
cd openclaude
bun install
bun run build

bash

git clone https://node.gitlawb.com/z6MkqDnb7Siv3Cwj7pGJq4T5EsUisECqR8KpnDLwcaZq5TPr/openclaude.git
cd openclaude
bun install
bun run build

optionally link globally

可选：全局链接

npm link

undefined

npm link

undefined

Run without build

无需构建直接运行

bash

bun run dev       # run directly with Bun, no build step

bash

bun run dev       # 直接通过Bun运行，无需构建步骤

Activation — Required Environment Variables

激活 — 必选环境变量

You must set

CLAUDE_CODE_USE_OPENAI=1

to enable the shim. Without it, the tool falls back to the Anthropic SDK.

Variable	Required	Purpose
`CLAUDE_CODE_USE_OPENAI`	Yes	Set to `1` to activate OpenAI provider
`OPENAI_API_KEY`	Yes*	API key (*omit for local Ollama/LM Studio)
`OPENAI_MODEL`	Yes	Model identifier
`OPENAI_BASE_URL`	No	Custom endpoint (default: `https://api.openai.com/v1` )
`CODEX_API_KEY`	Codex only	ChatGPT/Codex access token
`CODEX_AUTH_JSON_PATH`	Codex only	Path to Codex CLI `auth.json`

OPENAI_MODEL

takes priority over

ANTHROPIC_MODEL

if both are set.

你必须设置

CLAUDE_CODE_USE_OPENAI=1

来启用垫片，未设置的话工具会回退使用Anthropic SDK。

变量	必填	用途
`CLAUDE_CODE_USE_OPENAI`	是	设为 `1` 即可激活OpenAI提供商
`OPENAI_API_KEY`	是*	API密钥（*本地使用Ollama/LM Studio时可省略）
`OPENAI_MODEL`	是	模型标识符
`OPENAI_BASE_URL`	否	自定义接口端点（默认： `https://api.openai.com/v1` ）
`CODEX_API_KEY`	仅Codex需要	ChatGPT/Codex访问令牌
`CODEX_AUTH_JSON_PATH`	仅Codex需要	Codex CLI `auth.json` 的文件路径

如果同时设置了

OPENAI_MODEL

和

ANTHROPIC_MODEL

，

OPENAI_MODEL

优先级更高。

Provider Configuration Examples

提供商配置示例

OpenAI

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=$OPENAI_API_KEY
export OPENAI_MODEL=gpt-4o
openclaude

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=$OPENAI_API_KEY
export OPENAI_MODEL=gpt-4o
openclaude

DeepSeek

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=$DEEPSEEK_API_KEY
export OPENAI_BASE_URL=https://api.deepseek.com/v1
export OPENAI_MODEL=deepseek-chat
openclaude

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=$DEEPSEEK_API_KEY
export OPENAI_BASE_URL=https://api.deepseek.com/v1
export OPENAI_MODEL=deepseek-chat
openclaude

Google Gemini (via OpenRouter)

Google Gemini（通过OpenRouter接入）

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=$OPENROUTER_API_KEY
export OPENAI_BASE_URL=https://openrouter.ai/api/v1
export OPENAI_MODEL=google/gemini-2.0-flash
openclaude

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=$OPENROUTER_API_KEY
export OPENAI_BASE_URL=https://openrouter.ai/api/v1
export OPENAI_MODEL=google/gemini-2.0-flash
openclaude

Ollama (local, no API key needed)

Ollama（本地运行，无需API密钥）

bash

ollama pull llama3.3:70b

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_BASE_URL=http://localhost:11434/v1
export OPENAI_MODEL=llama3.3:70b
openclaude

bash

ollama pull llama3.3:70b

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_BASE_URL=http://localhost:11434/v1
export OPENAI_MODEL=llama3.3:70b
openclaude

Groq

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=$GROQ_API_KEY
export OPENAI_BASE_URL=https://api.groq.com/openai/v1
export OPENAI_MODEL=llama-3.3-70b-versatile
openclaude

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=$GROQ_API_KEY
export OPENAI_BASE_URL=https://api.groq.com/openai/v1
export OPENAI_MODEL=llama-3.3-70b-versatile
openclaude

Mistral

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=$MISTRAL_API_KEY
export OPENAI_BASE_URL=https://api.mistral.ai/v1
export OPENAI_MODEL=mistral-large-latest
openclaude

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=$MISTRAL_API_KEY
export OPENAI_BASE_URL=https://api.mistral.ai/v1
export OPENAI_MODEL=mistral-large-latest
openclaude

Azure OpenAI

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=$AZURE_OPENAI_KEY
export OPENAI_BASE_URL=https://your-resource.openai.azure.com/openai/deployments/your-deployment/v1
export OPENAI_MODEL=gpt-4o
openclaude

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=$AZURE_OPENAI_KEY
export OPENAI_BASE_URL=https://your-resource.openai.azure.com/openai/deployments/your-deployment/v1
export OPENAI_MODEL=gpt-4o
openclaude

Codex (ChatGPT backend)

Codex（ChatGPT后端）

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_MODEL=codexplan   # or codexspark for faster loops

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_MODEL=codexplan   # 更快的循环可使用codexspark

reads ~/.codex/auth.json automatically if present

如果存在~/.codex/auth.json会自动读取

or set: export CODEX_API_KEY=$CODEX_TOKEN

也可手动设置：export CODEX_API_KEY=$CODEX_TOKEN

openclaude

undefined

openclaude

undefined

LM Studio (local)

LM Studio（本地运行）

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_BASE_URL=http://localhost:1234/v1
export OPENAI_MODEL=your-model-name
openclaude

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_BASE_URL=http://localhost:1234/v1
export OPENAI_MODEL=your-model-name
openclaude

Together AI

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=$TOGETHER_API_KEY
export OPENAI_BASE_URL=https://api.together.xyz/v1
export OPENAI_MODEL=meta-llama/Llama-3.3-70B-Instruct-Turbo
openclaude

bash

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=$TOGETHER_API_KEY
export OPENAI_BASE_URL=https://api.together.xyz/v1
export OPENAI_MODEL=meta-llama/Llama-3.3-70B-Instruct-Turbo
openclaude

Architecture — How the Shim Works

架构 — 垫片工作原理

The shim file is

src/services/api/openaiShim.ts

(724 lines). It duck-types the Anthropic SDK interface so the rest of Claude Code is unaware it's talking to a different provider.

Claude Code Tool System
        │
        ▼
Anthropic SDK interface (duck-typed)
        │
        ▼
openaiShim.ts  ← format translation layer
        │
        ▼
OpenAI Chat Completions API
        │
        ▼
Any compatible model

垫片文件为

src/services/api/openaiShim.ts

（共724行），它通过鸭子类型实现了Anthropic SDK的接口，因此Claude Code的其余部分完全感知不到它正在和不同的提供商交互。

Claude Code 工具系统
        │
        ▼
Anthropic SDK 接口（鸭子类型实现）
        │
        ▼
openaiShim.ts  ← 格式转换层
        │
        ▼
OpenAI 聊天补全API
        │
        ▼
任意兼容模型

What the shim translates

垫片的转换内容

Anthropic message content blocks → OpenAI
```
messages
```
array
Anthropic
```
tool_use
```
/
```
tool_result
```
blocks → OpenAI
```
function_calls
```
/
```
tool
```
messages
OpenAI SSE streaming chunks → Anthropic stream events
Anthropic system prompt arrays → OpenAI
```
system
```
role messages

Anthropic消息内容块 → OpenAI
```
messages
```
数组
Anthropic
```
tool_use
```
/
```
tool_result
```
块 → OpenAI
```
function_calls
```
/
```
tool
```
消息
OpenAI SSE流块 → Anthropic流事件
Anthropic系统提示词数组 → OpenAI
```
system
```
角色消息

Files changed from upstream

相对于上游版本修改的文件

src/services/api/openaiShim.ts   ← NEW: the shim (724 lines)
src/services/api/client.ts       ← routes to shim when CLAUDE_CODE_USE_OPENAI=1
src/utils/model/providers.ts     ← added 'openai' provider type
src/utils/model/configs.ts       ← added openai model mappings
src/utils/model/model.ts         ← respects OPENAI_MODEL for defaults
src/utils/auth.ts                ← recognizes OpenAI as valid 3rd-party provider

src/services/api/openaiShim.ts   ← 新增：垫片文件（724行）
src/services/api/client.ts       ← 当CLAUDE_CODE_USE_OPENAI=1时路由到垫片
src/utils/model/providers.ts     ← 新增'openai'提供商类型
src/utils/model/configs.ts       ← 新增openai模型映射
src/utils/model/model.ts         ← 默认值优先读取OPENAI_MODEL
src/utils/auth.ts                ← 识别OpenAI为合法第三方提供商

Developer Workflow — Key Commands

开发者工作流 — 核心命令

bash

undefined

bash

undefined

Run in dev mode (no build)

开发模式运行（无需构建）

bun run dev

Build distribution

构建发布版本

bun run build

Launch with persisted profile (.openclaude-profile.json)

使用持久化配置文件启动（.openclaude-profile.json）

bun run dev:profile

Launch with OpenAI profile (requires OPENAI_API_KEY in shell)

使用OpenAI配置启动（需要shell中已设置OPENAI_API_KEY）

bun run dev:openai

Launch with Ollama profile (localhost:11434, llama3.1:8b default)

使用Ollama配置启动（默认localhost:11434，llama3.1:8b模型）

bun run dev:ollama

Launch with Codex profile

使用Codex配置启动

bun run dev:codex

Quick startup sanity check

快速启动可用性检查

bun run smoke

Validate provider env + reachability

验证提供商环境变量与可达性

bun run doctor:runtime

Machine-readable runtime diagnostics

机器可读的运行时诊断信息

bun run doctor:runtime:json

Persist diagnostics report to reports/doctor-runtime.json

将诊断报告持久化保存到reports/doctor-runtime.json

bun run doctor:report

Full local hardening check (typecheck + smoke + runtime doctor)

完整本地加固检查（类型检查+可用性检查+运行时诊断）

bun run hardening:check

Strict hardening (includes project-wide typecheck)

严格加固检查（包含全项目类型检查）

bun run hardening:strict

---

bun run hardening:strict

---

Profile Bootstrap — One-Time Setup

配置文件引导 — 一次性设置

Profiles save provider config to

.openclaude-profile.json

so you don't repeat env exports.

bash

undefined

配置文件会将提供商配置保存到

.openclaude-profile.json

，无需重复导出环境变量。

bash

undefined

Auto-detect provider (ollama if running, otherwise openai)

自动检测提供商（如果Ollama正在运行则用Ollama，否则用OpenAI）

bun run profile:init

Bootstrap for OpenAI

为OpenAI生成配置

bun run profile:init -- --provider openai --api-key $OPENAI_API_KEY

Bootstrap for Ollama with custom model

为使用自定义模型的Ollama生成配置

bun run profile:init -- --provider ollama --model llama3.1:8b

Bootstrap for Codex

为Codex生成配置

bun run profile:init -- --provider codex --model codexspark bun run profile:codex


After bootstrapping, run the app via the persisted profile:

```bash
bun run dev:profile

bun run profile:init -- --provider codex --model codexspark bun run profile:codex


引导完成后，即可通过持久化配置文件运行应用：

```bash
bun run dev:profile

TypeScript Integration — Using the Shim Directly

TypeScript集成 — 直接使用垫片

If you want to use the shim in your own TypeScript code:

typescript

// src/services/api/client.ts pattern — routing to the shim
import { openaiShim } from './openaiShim.js';

const useOpenAI = process.env.CLAUDE_CODE_USE_OPENAI === '1';

const client = useOpenAI
  ? openaiShim({
      apiKey: process.env.OPENAI_API_KEY,
      baseURL: process.env.OPENAI_BASE_URL ?? 'https://api.openai.com/v1',
      model: process.env.OPENAI_MODEL ?? 'gpt-4o',
    })
  : new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY });

typescript

// Streaming usage pattern (mirrors Anthropic SDK interface)
const stream = await client.messages.stream({
  model: process.env.OPENAI_MODEL!,
  max_tokens: 32000,
  system: 'You are a helpful coding assistant.',
  messages: [
    { role: 'user', content: 'Refactor this function for readability.' }
  ],
  tools: myTools, // Anthropic-format tool definitions — shim translates them
});

for await (const event of stream) {
  // events arrive in Anthropic format regardless of underlying provider
  if (event.type === 'content_block_delta') {
    process.stdout.write(event.delta.text ?? '');
  }
}

如果你想在自己的TypeScript代码中使用垫片：

typescript

// src/services/api/client.ts 模式 — 路由到垫片
import { openaiShim } from './openaiShim.js';

const useOpenAI = process.env.CLAUDE_CODE_USE_OPENAI === '1';

const client = useOpenAI
  ? openaiShim({
      apiKey: process.env.OPENAI_API_KEY,
      baseURL: process.env.OPENAI_BASE_URL ?? 'https://api.openai.com/v1',
      model: process.env.OPENAI_MODEL ?? 'gpt-4o',
    })
  : new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY });

typescript

// 流式使用模式（与Anthropic SDK接口完全一致）
const stream = await client.messages.stream({
  model: process.env.OPENAI_MODEL!,
  max_tokens: 32000,
  system: 'You are a helpful coding assistant.',
  messages: [
    { role: 'user', content: 'Refactor this function for readability.' }
  ],
  tools: myTools, // Anthropic格式的工具定义 — 垫片会自动转换
});

for await (const event of stream) {
  // 无论底层提供商是什么，事件都以Anthropic格式返回
  if (event.type === 'content_block_delta') {
    process.stdout.write(event.delta.text ?? '');
  }
}

Model Quality Reference

模型质量参考

Model	Tool Calling	Code Quality	Speed
GPT-4o	Excellent	Excellent	Fast
DeepSeek-V3	Great	Great	Fast
Gemini 2.0 Flash	Great	Good	Very Fast
Llama 3.3 70B	Good	Good	Medium
Mistral Large	Good	Good	Fast
GPT-4o-mini	Good	Good	Very Fast
Qwen 2.5 72B	Good	Good	Medium
Models < 7B	Limited	Limited	Very Fast

For agentic multi-step tool use, prefer models with strong native function/tool calling (GPT-4o, DeepSeek-V3, Gemini 2.0 Flash).

模型	Tool Calling	代码质量	速度
GPT-4o	优秀	优秀	快
DeepSeek-V3	很好	很好	快
Gemini 2.0 Flash	很好	好	非常快
Llama 3.3 70B	好	好	中等
Mistral Large	好	好	快
GPT-4o-mini	好	好	非常快
Qwen 2.5 72B	好	好	中等
参数小于7B的模型	有限	有限	非常快

对于多步骤Agent工具调用场景，优先选择原生函数/工具调用能力强的模型（GPT-4o、DeepSeek-V3、Gemini 2.0 Flash）。

What Works vs. What Doesn't

支持与不支持的功能

Fully supported

完全支持

All tools: Bash, FileRead, FileWrite, FileEdit, Glob, Grep, WebFetch, WebSearch, Agent, MCP, LSP, NotebookEdit, Tasks
Streaming (real-time token output)
Multi-step tool chains
Vision/images (base64 and URL) for models that support them
Slash commands:
```
/commit
```
,
```
/review
```
,
```
/compact
```
,
```
/diff
```
,
```
/doctor
```
Sub-agents (AgentTool spawns sub-agents using the same provider)
Persistent memory

所有工具：Bash、FileRead、FileWrite、FileEdit、Glob、Grep、WebFetch、WebSearch、Agent、MCP、LSP、NotebookEdit、Tasks
流式输出（实时token输出）
多步骤工具链
支持视觉/图片能力的模型可使用base64和URL格式的图片输入
斜杠命令：
```
/commit
```
、
```
/review
```
、
```
/compact
```
、
```
/diff
```
、
```
/doctor
```
子Agent（AgentTool使用相同提供商启动子Agent）
持久化记忆

Not supported (Anthropic-specific features)

不支持（Anthropic专属特性）

Extended thinking / reasoning mode
Prompt caching (Anthropic cache headers skipped)
Anthropic beta feature headers
Token output defaults to 32K max (gracefully capped if model is lower)

扩展思考/推理模式
提示词缓存（跳过Anthropic缓存头）
Anthropic beta功能头
Token输出默认最高32K（如果模型上限更低会自动适配）

Troubleshooting

问题排查

doctor:runtime

fails with placeholder key error

doctor:runtime

报错提示占位符密钥错误

Error: OPENAI_API_KEY looks like a placeholder (SUA_CHAVE)

Set a real key:

export OPENAI_API_KEY=$YOUR_ACTUAL_KEY

Error: OPENAI_API_KEY looks like a placeholder (SUA_CHAVE)

设置真实的密钥：

export OPENAI_API_KEY=$YOUR_ACTUAL_KEY

Ollama connection refused

Ollama连接被拒绝

Ensure Ollama is running before launching:

bash

ollama serve &
ollama pull llama3.3:70b
bun run dev:ollama

启动前确保Ollama正在运行：

bash

ollama serve &
ollama pull llama3.3:70b
bun run dev:ollama

Tool calls not working / model ignores tools

工具调用不工作/模型忽略工具

Switch to a model with strong tool calling support (GPT-4o, DeepSeek-V3). Models under 7B parameters often fail at multi-step agentic tool use.

切换到工具调用能力强的模型（GPT-4o、DeepSeek-V3），7B参数以下的模型通常无法胜任多步骤Agent工具调用任务。

Azure endpoint format

Azure端点格式错误

The

OPENAI_BASE_URL

for Azure must include the deployment path:

https://<resource>.openai.azure.com/openai/deployments/<deployment>/v1

Azure的

OPENAI_BASE_URL

必须包含部署路径：

https://<resource>.openai.azure.com/openai/deployments/<deployment>/v1

Codex auth not found

找不到Codex认证信息

~/.codex/auth.json

doesn't exist, set the token directly:

bash

export CODEX_API_KEY=$YOUR_CODEX_TOKEN

Or point to a custom auth file:

bash

export CODEX_AUTH_JSON_PATH=/path/to/auth.json

如果

~/.codex/auth.json

不存在，可直接设置令牌：

bash

export CODEX_API_KEY=$YOUR_CODEX_TOKEN

或者指向自定义的认证文件：

bash

export CODEX_AUTH_JSON_PATH=/path/to/auth.json

Run diagnostics for any issue

任何问题都可运行诊断命令

bash

bun run doctor:runtime       # human-readable
bun run doctor:runtime:json  # machine-readable JSON
bun run doctor:report        # saves to reports/doctor-runtime.json

bash

bun run doctor:runtime       # 人类可读格式
bun run doctor:runtime:json  # 机器可读JSON格式
bun run doctor:report        # 保存到reports/doctor-runtime.json