Agentify Your App with GitHub Copilot’s Agentic Coding SDK
import asyncio
import sys
from copilot import CopilotClient
from copilot.tools import define_tool
from copilot.generated.session_events import SessionEventType
from pydantic import BaseModel, Field
# Step 1: Define custom tools using the @define_tool decorator.
class GetDataVisualizationParams(BaseModel):
library_name: str = Field(description=“The name of the Python library to get info about”)
@define_tool(description=“Get information about a Python data visualization library”)
async def get_library_info(params: GetDataVisualizationParams) -> dict:
“”“Custom tool that provides information about data visualization libraries.”“”
libraries = {
“matplotlib”: {
“name”: “Matplotlib”,
“use_case”: “Foundational plotting library for static, animated, and interactive visualizations”,
“install”: “pip install matplotlib”,
“popularity”: “Most widely used, basis for many other libraries”,
},
“seaborn”: {
“name”: “Seaborn”,
“use_case”: “Statistical data visualization with attractive default styles”,
“install”: “pip install seaborn”,
“popularity”: “Great for exploratory data analysis”,
},
“plotly”: {
“name”: “Plotly”,
“use_case”: “Interactive, publication-quality graphs for dashboards”,
“install”: “pip install plotly”,
“popularity”: “Best for web-based interactive visualizations”,
},
}
library = params.library_name.lower()
if library in libraries:
return libraries[library]
return {“error”: f“Library ‘{library}’ not found. Try: matplotlib, seaborn, or plotly”}
async def main():
# Step 2: Create and start the Copilot client with an explicit CLI path.
# The SDK needs to find the Copilot CLI, so specify the path explicitly.
client = CopilotClient({
“cli_path”: “C:\\nvm4w\\nodejs\\copilot.cmd”, # Path to Copilot CLI
“log_level”: “debug”, # Enable debug logging for troubleshooting
})
print(” GitHub Copilot SDK Demo – Agentic Coding in Action”)
print(” Starting Copilot client (this may take a moment)…\n”)
await client.start()
print(“=” * 60)
# Step 3: Create a session with custom configuration.
session = await client.create_session({
“model”: “gpt-4.1”, # Choose a model
“streaming”: True, # Enable streaming responses
“tools”: [get_library_info], # Register custom tools
“system_message”: (
“You are a helpful technical assistant for data scientists. “
“When asked about visualization libraries, use the get_library_info tool “
“to provide accurate information.”
),
})
print(f“Session created: {session.session_id}\n”)
# Step 4: Set up event handlers for streaming.
def handle_event(event):
if event.type == SessionEventType.ASSISTANT_MESSAGE_DELTA:
# Stream the response as it arrives.
sys.stdout.write(event.data.delta_content)
sys.stdout.flush()
elif event.type == SessionEventType.TOOL_EXECUTION_START:
print(f“\n Tool called: {event.data.tool_name}”)
session.on(handle_event)
# Step 5: Send a prompt and let the agent work.
print(” User: List three common Python libraries for data visualization and their main use case.\n”)
print(” Assistant: “, end=“”)
await session.send_and_wait({
“prompt”: (
“List three common Python libraries for data visualization and their main use case. “
“Use the get_library_info tool to get accurate information about each one.”
)
})
print(“\n\n” + “=” * 60)
# Step 6: Clean up.
await session.destroy()
await client.stop()
print(” Session ended successfully!”)
if __name__ == “__main__”:
asyncio.run(main())
