Introduction
Adapters make DSPy modular and extensive. They handle communication with external resources like databases, APIs, and file systems, ensuring your core logic remains clean.
Built-in vs. Custom Adapters
Database Adapter (Example)
A typical PostgreSQL adapter pattern:
Python
class PostgreSQLAdapter(dspy.Adapter):
def __init__(self, connection_string):
super().__init__()
self.connection_string = connection_string
def query(self, sql, params=None):
# Implementation of query execution...
passCreating Custom Adapters
File System Adapter
Manage file operations seamlessly within your DSPy pipeline:
Python
import dspy
from pathlib import Path
class FileSystemAdapter(dspy.Adapter):
def __init__(self, base_path="."):
super().__init__()
self.base_path = Path(base_path)
def read_file(self, filename):
return (self.base_path / filename).read_text()
def write_file(self, filename, content):
(self.base_path / filename).write_text(content)API Adapter
Interact with external REST APIs:
Python
import requests
class APIAdapter(dspy.Adapter):
def __init__(self, base_url):
super().__init__()
self.base_url = base_url
def get(self, endpoint):
return requests.get(f"{self.base_url}/{endpoint}").json()Specialized Tools
Tools encapsulate specific functionality like calculations or text processing.
Calculator Tool
Python
class CalculatorTool(dspy.Tool):
def calculate(self, expression):
# Safe evaluation logic...
passIntegration Example: Tool-Enabled Agent
Combining adapters and tools in a powerful agent:
Python
class ToolEnabledAgent(dspy.Module):
def __init__(self):
super().__init__()
self.tools = {
'calculator': CalculatorTool(),
'file_system': FileSystemAdapter()
}
self.decide = dspy.Predict("task -> tool, params")
def forward(self, task):
decision = self.decide(task=task)
# Execute selected tool...
pass