Multi-Model AI Coding Agent CLI – OpenClaude

OpenClaude is an open source coding agent CLI that runs in the terminal and connects one workflow to cloud APIs and local model backends.

The coding agent executes prompts, file operations, and multi-step agent tasks directly from the command line. You can use it to maintain a single coding environment while switching between providers like OpenAI, Gemini, DeepSeek, and local Ollama models.

It supports OpenAI compatible endpoints, Gemini, GitHub Models, Codex, Ollama, Atomic Chat, and several environment based enterprise providers. You run local models on your own hardware or connect to enterprise cloud APIs based on your specific project requirements.

Features

• Executes bash commands, file edits, grep searches, and glob patterns directly in the terminal.
• Streams real-time token output and tool progress during execution.
• Processes multi-step tool loops with model calls and follow-up responses.
• Accepts URL and base64 image inputs for vision-capable models.
• Saves provider configurations in a local profile file for quick switching.
• Directs specific agent tasks to different models based on user settings.
• Searches the internet using DuckDuckGo or Firecrawl to retrieve external information.
• Runs as a headless gRPC service for integration into CI/CD pipelines or custom interfaces.
• Integrates with VS Code for launch control and theme support.

How to Use It

Getting Started

1. Install OpenClaude as a global CLI package.

npm install -g @gitlawb/openclaude

2. The package expects ripgrep to exist on the system path. Check that first if the CLI reports a missing dependency.

rg --version

3. Start the CLI with the default entry command.

openclaude

4. Use the built in provider flow after launch if you want guided setup and saved profiles.

/provider

5. Use the GitHub Models onboarding flow if that is your target backend.

/onboard-github

6. OpenAI setup uses OpenAI compatible environment variables plus a provider switch.

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=sk-your-key-here
export OPENAI_MODEL=gpt-4o
openclaude

7. Windows PowerShell uses the same variables in PowerShell syntax.

$env:CLAUDE_CODE_USE_OPENAI="1"
$env:OPENAI_API_KEY="sk-your-key-here"
$env:OPENAI_MODEL="gpt-4o"
openclaude

8. Local Ollama setup points the same OpenAI compatible interface at the Ollama server.

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_BASE_URL=http://localhost:11434/v1
export OPENAI_MODEL=qwen2.5-coder:7b
openclaude

9. PowerShell syntax for the same local setup looks like this.

$env:CLAUDE_CODE_USE_OPENAI="1"
$env:OPENAI_BASE_URL="http://localhost:11434/v1"
$env:OPENAI_MODEL="qwen2.5-coder:7b"
openclaude

Provider setup

Available Commands

Environment variables

Agent routing example

OpenClaude can map different agent roles to different backends through ~/.claude/settings.json.

{
  "agentModels": {
    "deepseek-chat": {
      "base_url": "https://api.deepseek.com/v1",
      "api_key": "sk-your-key"
    },   "gpt-4o": {
      "base_url": "https://api.openai.com/v1",
      "api_key": "sk-your-key"
    }
  },
  "agentRouting": {
    "Explore": "deepseek-chat",
    "Plan": "gpt-4o",
    "general-purpose": "gpt-4o",
    "frontend-dev": "deepseek-chat",
    "default": "gpt-4o"
  }
}

Web tool behavior

WebSearch uses DuckDuckGo by default on non Anthropic models. WebFetch uses a basic HTTP plus HTML to markdown path unless Firecrawl is configured.

# Set a Firecrawl API key 
export FIRECRAWL_API_KEY=your-key-here

Headless server mode

The gRPC server listens on localhost:50051 by default and streams text chunks, tool calls, and permission requests over a bidirectional connection. The repository also includes a CLI client and a Protocol Buffers definition file for generating clients in other languages.

# Start the gRPC Server 
npm run dev:grpc

# Run the Test CLI Client
npm run dev:grpc:cli

Pros

• Works with over 200 models through OpenAI-compatible endpoints plus native provider integrations.
• Runs local inference for free via Ollama or Atomic Chat.
• Routes different agents to different models to optimize cost and performance.
• Provides a headless gRPC mode for programmatic integration into CI/CD and external applications.
• Includes a VS Code extension for launch integration and theme support.

Cons

• The application stores API keys in plaintext within the settings file.
• Smaller local models struggle with long multi-step tool flows.
• Non-Anthropic providers lack Anthropic-specific features.
• The default DuckDuckGo search faces potential rate limits or blocks.

Related Resources

• ripgrep: Install the required search tool.
• Ollama: Run local models through an OpenAI-compatible endpoint.
• Firecrawl: Add stronger search and fetch behavior for pages that need JavaScript rendering.
• Best CLI AI Coding Agents: Discover more CLI AI coding agents.