openclaude-multi-llm

$27

git clone https://node.gitlawb.com/z6MkqDnb7Siv3Cwj7pGJq4T5EsUisECqR8KpnDLwcaZq5TPr/openclaude.git

cd openclaude

bun install

bun run build

# optionally link globally

npm link

Run without build

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

Activation — Required Environment Variables

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.

Provider Configuration Examples

OpenAI

export CLAUDE_CODE_USE_OPENAI=1

export OPENAI_API_KEY=$OPENAI_API_KEY

export OPENAI_MODEL=gpt-4o

openclaude

DeepSeek

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)

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 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

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

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

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)

export CLAUDE_CODE_USE_OPENAI=1

export OPENAI_MODEL=codexplan   # or codexspark for faster loops

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

# or set: export CODEX_API_KEY=$CODEX_TOKEN

openclaude

LM Studio (local)

export CLAUDE_CODE_USE_OPENAI=1

export OPENAI_BASE_URL=http://localhost:1234/v1

export OPENAI_MODEL=your-model-name

openclaude

Together AI

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

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

What the shim translates

• 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

Files changed from upstream

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

Developer Workflow — Key Commands

# Run in dev mode (no build)

bun run dev

# Build distribution

bun run build

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

bun run dev:profile

# Launch with OpenAI profile (requires OPENAI_API_KEY in shell)

bun run dev:openai

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

bun run dev:ollama

# Launch with Codex profile

bun run dev:codex

# Quick startup sanity check

bun run smoke

# Validate provider env + reachability

bun run doctor:runtime

# Machine-readable runtime diagnostics

bun run doctor:runtime:json

# Persist diagnostics report to reports/doctor-runtime.json

bun run doctor:report

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

bun run hardening:check

# Strict hardening (includes project-wide typecheck)

bun run hardening:strict

Profile Bootstrap — One-Time Setup

Profiles save provider config to .openclaude-profile.json so you don't repeat env exports.

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

bun run profile:init

# Bootstrap for OpenAI

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

# Bootstrap for Ollama with custom model

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

# Bootstrap for Codex

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

bun run profile:codex

After bootstrapping, run the app via the persisted profile:

bun run dev:profile

TypeScript Integration — Using the Shim Directly

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

// 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 });

// 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 ?? '');

  }

}

Model Quality Reference

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).

What Works vs. What Doesn't

Fully supported

• 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

Not supported (Anthropic-specific features)

• 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)

Troubleshooting

doctor:runtime fails with placeholder key error

Error: OPENAI_API_KEY looks like a placeholder (SUA_CHAVE)

Set a real key: export OPENAI_API_KEY=$YOUR_ACTUAL_KEY

Ollama connection refused

Ensure Ollama is running before launching:

ollama serve &

ollama pull llama3.3:70b

bun run dev:ollama

Tool calls not working / model ignores tools

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.

Azure endpoint format

The OPENAI_BASE_URL for Azure must include the deployment path:

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

Codex auth not found

If ~/.codex/auth.json doesn't exist, set the token directly:

export CODEX_API_KEY=$YOUR_CODEX_TOKEN

Or point to a custom auth file:

export CODEX_AUTH_JSON_PATH=/path/to/auth.json

Run diagnostics for any issue

bun run doctor:runtime       # human-readable

bun run doctor:runtime:json  # machine-readable JSON

bun run doctor:report        # saves to reports/doctor-runtime.json