Skip to content

Latest commit

 

History

History
239 lines (181 loc) · 7.61 KB

langgraph-integration.md

File metadata and controls

239 lines (181 loc) · 7.61 KB

LangGraph Integration Guide

This guide explains how to use the zeroclaw-tools Python package for consistent tool calling with any OpenAI-compatible LLM provider.

Background

Some LLM providers, particularly Chinese models like GLM-5 (Zhipu AI), have inconsistent tool calling behavior when using text-based tool invocation. ZeroClaw's Rust core uses structured tool calling via the OpenAI API format, but some models respond better to a different approach.

LangGraph provides a stateful graph execution engine that guarantees consistent tool calling behavior regardless of the underlying model's native capabilities.

Architecture

┌─────────────────────────────────────────────────────────────┐
│                      Your Application                        │
├─────────────────────────────────────────────────────────────┤
│                   zeroclaw-tools Agent                       │
│                                                              │
│   ┌─────────────────────────────────────────────────────┐   │
│   │              LangGraph StateGraph                    │   │
│   │                                                      │   │
│   │    ┌────────────┐         ┌────────────┐            │   │
│   │    │   Agent    │ ──────▶ │   Tools    │            │   │
│   │    │   Node     │ ◀────── │   Node     │            │   │
│   │    └────────────┘         └────────────┘            │   │
│   │         │                       │                    │   │
│   │         ▼                       ▼                    │   │
│   │    [Continue?]            [Execute Tool]             │   │
│   │         │                       │                    │   │
│   │    Yes │ No                Result│                    │   │
│   │         ▼                       ▼                    │   │
│   │      [END]              [Back to Agent]              │   │
│   │                                                      │   │
│   └─────────────────────────────────────────────────────┘   │
│                                                              │
├─────────────────────────────────────────────────────────────┤
│            OpenAI-Compatible LLM Provider                    │
│   (Z.AI, OpenRouter, Groq, DeepSeek, Ollama, etc.)          │
└─────────────────────────────────────────────────────────────┘

Quick Start

Installation

pip install zeroclaw-tools

Basic Usage

import asyncio
from zeroclaw_tools import create_agent, shell, file_read, file_write
from langchain_core.messages import HumanMessage

async def main():
    agent = create_agent(
        tools=[shell, file_read, file_write],
        model="glm-5",
        api_key="your-api-key",
        base_url="https://api.z.ai/api/coding/paas/v4"
    )
    
    result = await agent.ainvoke({
        "messages": [HumanMessage(content="Read /etc/hostname and tell me the machine name")]
    })
    
    print(result["messages"][-1].content)

asyncio.run(main())

Available Tools

Core Tools

Tool Description
shell Execute shell commands
file_read Read file contents
file_write Write content to files

Extended Tools

Tool Description
web_search Search the web (requires BRAVE_API_KEY)
http_request Make HTTP requests
memory_store Store data in persistent memory
memory_recall Recall stored data

Custom Tools

Create your own tools with the @tool decorator:

from zeroclaw_tools import tool, create_agent

@tool
def get_weather(city: str) -> str:
    """Get the current weather for a city."""
    # Your implementation
    return f"Weather in {city}: Sunny, 25°C"

@tool
def query_database(sql: str) -> str:
    """Execute a SQL query and return results."""
    # Your implementation
    return "Query returned 5 rows"

agent = create_agent(
    tools=[get_weather, query_database],
    model="glm-5",
    api_key="your-key"
)

Provider Configuration

Z.AI / GLM-5

agent = create_agent(
    model="glm-5",
    api_key="your-zhipu-key",
    base_url="https://api.z.ai/api/coding/paas/v4"
)

OpenRouter

agent = create_agent(
    model="anthropic/claude-sonnet-4-6",
    api_key="your-openrouter-key",
    base_url="https://openrouter.ai/api/v1"
)

Groq

agent = create_agent(
    model="llama-3.3-70b-versatile",
    api_key="your-groq-key",
    base_url="https://api.groq.com/openai/v1"
)

Ollama (Local)

agent = create_agent(
    model="llama3.2",
    base_url="http://localhost:11434/v1"
)

Discord Bot Integration

import os
from zeroclaw_tools.integrations import DiscordBot

bot = DiscordBot(
    token=os.environ["DISCORD_TOKEN"],
    guild_id=123456789,  # Your Discord server ID
    allowed_users=["123456789"],  # User IDs that can use the bot
    api_key=os.environ["API_KEY"],
    model="glm-5"
)

bot.run()

CLI Usage

# Set environment variables
export API_KEY="your-key"
export BRAVE_API_KEY="your-brave-key"  # Optional, for web search

# Single message
zeroclaw-tools "What is the current date?"

# Interactive mode
zeroclaw-tools -i

Comparison with Rust ZeroClaw

Aspect Rust ZeroClaw zeroclaw-tools
Performance Ultra-fast (~10ms startup) Python startup (~500ms)
Memory <5 MB ~50 MB
Binary size ~3.4 MB pip package
Tool consistency Model-dependent LangGraph guarantees
Extensibility Rust traits Python decorators
Ecosystem Rust crates PyPI packages

When to use Rust ZeroClaw:

  • Production edge deployments
  • Resource-constrained environments (Raspberry Pi, etc.)
  • Maximum performance requirements

When to use zeroclaw-tools:

  • Models with inconsistent native tool calling
  • Python-centric development
  • Rapid prototyping
  • Integration with Python ML ecosystem

Troubleshooting

"API key required" error

Set the API_KEY environment variable or pass api_key to create_agent().

Tool calls not executing

Ensure your model supports function calling. Some older models may not support tools.

Rate limiting

Add delays between calls or implement your own rate limiting:

import asyncio

for message in messages:
    result = await agent.ainvoke({"messages": [message]})
    await asyncio.sleep(1)  # Rate limit

Related Projects