Skip to main content

​
Understanding Tool Discovery

PraisonAI automatically discovers tools from ~/.praison/tools/. Any .py file you place there will be loaded and its functions become available as tools.
1

Create Your Tools File

# my_pandas_tools.py
import pandas as pd
from io import StringIO

def analyze_csv_data(csv_data: str) -> str:
    """Analyze CSV data using pandas DataFrame.
    
    Args:
        csv_data: CSV-formatted string data
        
    Returns:
        Analysis summary
    """
    df = pd.read_csv(StringIO(csv_data))
    return f"""Shape: {df.shape}
Columns: {list(df.columns)}
Statistics:
{df.describe().to_string()}"""
2

Add File to Tools Directory

praisonai tools add ./my_pandas_tools.py
Output:
✅ Added tools file: my_pandas_tools.py
   Copied to: ~/.praison/tools/my_pandas_tools.py
   Found 1 tools: analyze_csv_data
3

Verify Tools Are Discovered

Tools in ~/.praison/tools/ are auto-discovered:
from praisonai.templates.tool_override import create_tool_registry_with_overrides

registry = create_tool_registry_with_overrides(include_defaults=True)
print("analyze_csv_data" in registry)  # True
4

Use in Agent

from praisonaiagents import Agent

agent = Agent(
    name="data_analyst",
    role="Data Analyst",
    tools=[analyze_csv_data]
)

csv = "name,age\\nAlice,30\\nBob,25"
result = agent.start(f"Analyze this data: {csv}")

​
Why Packages Need Wrapper Tools

Running praisonai tools add pandas will show: 
⚠️  Package 'pandas' is installed but NOT directly usable as tools
    Found 102 callable items in package

    To use this package with agents, create wrapper tools:
    1. Create a file: ~/.praison/tools/my_tools.py
    2. Define wrapper functions with docstrings
    3. Tools will be auto-discovered
Packages like pandas are NOT tools - they are libraries. To use them with agents, you must create wrapper functions with:
  • Type hints for parameters
  • Docstrings with Args section
  • JSON-serializable return values

​
How to Add Tools from Local Files

1

Create Tools File

# my_tools.py
def my_custom_tool(query: str) -> str:
    """Process a query.
    
    Args:
        query: Input query
        
    Returns:
        Processed result
    """
    return f"Processed: {query}"
2

Add File to Tools

praisonai tools add ./my_tools.py
Output:
✅ Added tools file: my_tools.py
   Copied to: ~/.praison/tools/my_tools.py
   Found 1 tools: my_custom_tool
3

Use in Template

Reference the tool in your agents.yaml:
roles:
  processor:
    tools:
      - my_custom_tool

​
How to Add Tools from GitHub

1

Add from GitHub Repository

praisonai tools add github:MervinPraison/PraisonAI-tools/praisonai_tools
2

Verify Download

Tools are downloaded to ~/.praison/tools/

​
Configuration File

Tools sources are stored in ~/.praison/tools_sources.yaml: 
sources:
- pandas
- numpy

​
Key Concept

Packages vs Tools: When you add a package like pandas, you’re registering it as a source. To use it with agents, you need to create wrapper functions that expose specific functionality as tools with proper docstrings and type hints.