OpenClaw llm-box Plugin

June 30, 2026 · View on GitHub

Invoke llm-box workflows as tools in OpenClaw conversations

Overview

This plugin integrates llm-box with OpenClaw, allowing you to invoke llm-box workflows as callable tools within OpenClaw conversations. Your AI assistant can now execute complex, multi-step workflows built with llm-box directly from chat.

Features

Workflow Discovery - List all available llm-box workflows in your configured directory
Workflow Execution - Execute any workflow by name with optional input parameters
Workflow Inspection - Get detailed descriptions of what each workflow does
Multi-Provider Support - Leverage 20+ LLM providers through llm-box (Ollama, Kimi, DeepSeek, Qwen, etc.)

How It Works

┌─────────────────────────────────────────────────────────────┐
│                        OpenClaw Agent                         │
│                                                               │
│  You: "Summarize this article using my kimi_summary workflow"│
│                                                               │
│  Agent → llmbox_list_workflows (tool)                        │
│         ↓                                                    │
│  Agent → llmbox_run_workflow(workflow_file="kimi_summary")   │
│         ↓                                                    │
│  llm-box executes:                                            │
│    [STEP 1] fetch_url → Extract article content              │
│    [STEP 2] kimi → Generate summary                          │
│         ↓                                                    │
│  Agent: "Here's the summary: ..."                            │
└─────────────────────────────────────────────────────────────┘

Prerequisites

OpenClaw installed and configured
llm-box installed (go install github.com/alib8b8/llm-box@latest)
Some llm-box workflows in your workflows directory

Installation

# Install the plugin
openclaw plugin install alib8b8/openclaw-llmbox

# Or clone and install locally
git clone https://github.com/alib8b8/openclaw-llmbox.git
cd openclaw-llmbox
npm install
npm run build
openclaw plugin install ./dist

Configuration

Edit your OpenClaw config file (~/.openclaw/config.json):

{
  "plugins": {
    "openclaw-llmbox": {
      "workflowDir": "./workflows",
      "llmboxPath": "llm-box",
      "enableAutoDiscovery": true
    }
  }
}

Config Option	Default	Description
`workflowDir`	`./workflows`	Directory containing your llm-box workflow YAML files
`llmboxPath`	`llm-box`	Path to llm-box binary (must be in PATH or use full path)
`enableAutoDiscovery`	`true`	Auto-discover workflows in the configured directory

Available Tools

`llmbox_list_workflows`

List all available llm-box workflows in your configured directory.

Parameters: None

Example:

User: What workflows do I have available?

Agent calls llmbox_list_workflows:
→ {"workflows": [
    {"name": "kimi_summary", "file": "kimi_summary.yaml", "steps": 2},
    {"name": "deepseek_coder", "file": "deepseek_coder.yaml", "steps": 3},
    {"name": "multi_step", "file": "multi_step.yaml", "steps": 5}
  ], "count": 3}

`llmbox_run_workflow`

Execute a specific llm-box workflow.

Parameters:

workflow_file (required): The workflow filename (e.g., "kimi_summary.yaml")
input (optional): Input text to pass to the workflow as {{input}}

Example:

User: Summarize https://example.com/article using kimi_summary

Agent calls llmbox_run_workflow:
→ {"workflow_file": "kimi_summary.yaml", "input": "https://example.com/article", ...}

`llmbox_describe_workflow`

Get detailed information about a specific workflow.

Parameters:

workflow_file (required): The workflow filename

Example:

User: What does the deepseek_coder workflow do?

Agent calls llmbox_describe_workflow:
→ {"workflow": "deepseek_coder.yaml", "name": "deepseek_coder", 
   "description": "Code review and fix workflow using DeepSeek Coder",
   "steps": [
     {"step": 1, "node": "fetch_url", "params": {"url": "{{input}}"}},
     {"step": 2, "node": "deepseek", "params": {"model": "deepseek-coder", "prompt": "Review this code..."}}
   ]}

Creating Workflows for OpenClaw

Here are some workflow patterns that work great with OpenClaw:

1. Simple URL Processing

# workflows/url_summarize.yaml
name: URL Summarizer
description: Fetch a URL and summarize its content
steps:
  - node: fetch_url
    params:
      url: "{{input}}"
  - node: deepseek
    params:
      model: deepseek-chat
      prompt: "Summarize the following content in 3 bullet points:\n\n{{output}}"

2. Multi-Provider Pipeline

# workflows/research_assistant.yaml
name: Research Assistant
description: Research a topic using multiple sources
steps:
  - node: fetch_url
    params:
      url: "{{input}}"
  - node: ollama
    params:
      model: llama3
      prompt: "Extract key information from: {{output}}"
  - node: kimi
    params:
      model: moonshot-v1-128k
      prompt: "Format the following research notes:\n\n{{output}}"

3. Code Review

# workflows/code_review.yaml
name: Code Review
description: Review code for bugs and improvements
steps:
  - node: file_read
    params:
      path: "{{input}}"
  - node: deepseek
    params:
      model: deepseek-coder
      prompt: "Review this code for:\n1. Bugs\n2. Security issues\n3. Performance improvements\n\nCode:\n{{output}}"

Example Usage

Chat Mode

$ openclaw chat

🤖: How can I help you today?

You: I want to summarize this article: https://example.com/tech-news

🤖: Let me use the kimi_summary workflow to summarize that article.

[Calling llmbox_run_workflow]
→ Workflow: kimi_summary.yaml
→ Steps executed: 2
→ Duration: 3.2s

🤖: Here's a summary of the article:

• **Main Topic**: The article discusses the latest developments in AI technology
• **Key Points**: 
  - New LLM models achieving state-of-the-art results
  - Increased focus on efficiency and cost reduction
  - More open-source releases
• **Conclusion**: The AI field continues to evolve rapidly...

Command Mode

$ openclaw "Summarize https://example.com/article using my kimi_summary workflow"

Requirements

Node.js >= 22.0.0
OpenClaw
llm-box

License

MIT - See LICENSE

Contributing

Contributions are welcome! Please open an issue or submit a PR.