openclaw-cli-providers

OpenClaw provider plugins that route LLM calls through locally installed CLI tools instead of calling cloud APIs directly. No API keys needed in OpenClaw — authentication is handled by each CLI tool.

Packages

| Package | CLI required | npm | |---|---|---| | @richizo/openclaw-claude-code-cli | Claude Code (claude) |  | | @richizo/openclaw-gemini-cli | Gemini CLI (gemini) |  | | @richizo/openclaw-qwen-code-cli | Qwen Code (qwen) |  |

Quick install

# Install one or more plugins
openclaw plugins install @richizo/openclaw-claude-code-cli
openclaw plugins install @richizo/openclaw-gemini-cli
openclaw plugins install @richizo/openclaw-qwen-code-cli

Then enable them in your openclaw.json:

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "claude-code-cli/claude-sonnet-4-6"
      },
      "models": {
        "claude-code-cli/claude-sonnet-4-6": {},
        "claude-code-cli/claude-opus-4-6": {},
        "gemini-cli/gemini-2.5-flash": {},
        "gemini-cli/gemini-2.5-pro": {},
        "qwen-code-cli/qwen-coder": {},
        "qwen-code-cli/qwen3-coder-plus": {}
      }
    }
  },
  "plugins": {
    "entries": {
      "claude-code-cli": { "enabled": true },
      "gemini-cli": { "enabled": true },
      "qwen-code-cli": { "enabled": true }
    }
  }
}

How it works

Each plugin spawns the corresponding CLI binary as a subprocess, passes the conversation as a formatted prompt, and streams the response back through OpenClaw's AssistantMessageEventStream protocol.

OpenClaw agent  →  plugin  →  CLI subprocess  →  LLM (via CLI's own auth)
                   ↑                              ↑
              no API key                  user's existing login
              in OpenClaw                  (OAuth, API key, etc.)

Prerequisites

• OpenClaw >= 2026.3.24
• Node.js >= 22
• The respective CLI binary installed and authenticated (see each package README)

License

MIT — see LICENSE