Microsoft Foundry Local को बड़े अनुप्रयोगों में कॉल करने योग्य टूल्स के रूप में एकीकृत करने के लिए एक व्यापक ढांचा, जो टूल-आधारित AI इंटीग्रेशन के लिए Microsoft के आधिकारिक पैटर्न का पालन करता है।
यह उदाहरण दिखाता है कि Foundry Local मॉडल्स को पुन: उपयोग योग्य टूल्स के रूप में कैसे प्रस्तुत किया जा सकता है, जिन्हें मौजूदा अनुप्रयोगों, वर्कफ़्लो और विकास परिवेशों में एकीकृत किया जा सकता है। यह टूल इंटीग्रेशन और फंक्शन कॉलिंग के लिए Microsoft द्वारा अनुशंसित पैटर्न को प्रदर्शित करता है।
- Foundry Local मॉडल्स को कॉल करने योग्य फंक्शन्स के रूप में उपयोग करना
- मानकीकृत टूल इंटरफेस और स्कीमाज
- मौजूदा कोडबेस के साथ सहज एकीकरण
- टाइप-सुरक्षित टूल परिभाषाएँ और सत्यापन
- Microsoft Foundry Local फंक्शन कॉलिंग का कार्यान्वयन
- OpenAI-संगत टूल परिभाषाएँ
- स्वचालित पैरामीटर सत्यापन और रूपांतरण
- त्रुटि प्रबंधन और प्रतिक्रिया स्वरूपण
- LangChain इंटीग्रेशन: नेटिव LangChain टूल समर्थन
- Semantic Kernel: Microsoft Semantic Kernel फंक्शन्स
- REST API: HTTP-आधारित टूल एंडपॉइंट्स
- CLI टूल्स: कमांड-लाइन इंटरफेस इंटीग्रेशन
- Jupyter Notebooks: इंटरैक्टिव विकास उपकरण
- कोड विश्लेषण और निर्माण टूल्स
- सामग्री प्रसंस्करण और सारांशण
- डेटा विश्लेषण और विज़ुअलाइज़ेशन
- शोध और जानकारी पुनर्प्राप्ति
- निर्णय समर्थन प्रणाली
┌─────────────────────────────────────────────────────────────────┐
│ Application Layer │
│ │
│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │
│ │ LangChain │ │ Semantic │ │ Custom │ │
│ │ Tools │ │ Kernel │ │ Apps │ │
│ └─────────────┘ └─────────────┘ └─────────────┘ │
└─────────────────┬─────────────────┬─────────────────────────────┘
│ │
▼ ▼
┌─────────────────────────────────────────────────────────────────┐
│ Tool Integration Layer │
│ │
│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │
│ │ Function │ │ REST │ │ CLI │ │
│ │ Registry │ │ Gateway │ │ Interface │ │
│ └─────────────┘ └─────────────┘ └─────────────┘ │
└─────────────────────────┬───────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ Microsoft Foundry Local Service │
│ │
│ • Model Management • Function Calling Support │
│ • Inference Engine • Tool Schema Validation │
│ • Context Handling • Response Formatting │
└─────────────────────────────────────────────────────────────────┘
- Python: 3.9+ (asyncio समर्थन के साथ)
- Node.js: v18+ (JavaScript इंटीग्रेशन के लिए)
- मेमोरी: 12GB+ अनुशंसित
- स्टोरेज: मॉडल्स और टूल्स के लिए 10GB+
pip install foundry-local-sdk openai langchain semantic-kernel fastapi uvicorn typer rich# LangChain integration
pip install langchain-openai langchain-community
# Semantic Kernel integration
pip install semantic-kernel
# Web framework integration
pip install fastapi uvicorn streamlit gradio
# Development tools
pip install jupyter ipywidgetsfrom foundry_tools import FoundryTool, FoundryToolRegistry
# Create a simple analysis tool
@FoundryTool(
name="code_analyzer",
description="Analyze code quality and suggest improvements",
model="phi-4-mini"
)
async def analyze_code(code: str, language: str = "python") -> dict:
"""Analyze code and return quality metrics and suggestions."""
pass
# Register and use the tool
registry = FoundryToolRegistry()
await registry.register(analyze_code)
result = await registry.call("code_analyzer", {
"code": "def hello(): print('world')",
"language": "python"
})from langchain.tools import BaseTool
from foundry_tools.langchain import FoundryLangChainTool
# Create LangChain-compatible tool
class CodeAnalyzerTool(FoundryLangChainTool):
name = "code_analyzer"
description = "Analyze code quality using Foundry Local"
model = "phi-4-mini"
async def _arun(self, code: str, language: str = "python") -> str:
return await self.foundry_call({
"code": code,
"language": language
})
# Use with LangChain agents
from langchain.agents import initialize_agent, AgentType
tools = [CodeAnalyzerTool()]
agent = initialize_agent(
tools=tools,
llm=None, # Uses Foundry Local
agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION
)from fastapi import FastAPI
from foundry_tools.rest import FoundryRESTEndpoint
app = FastAPI()
# Auto-generate REST endpoints from Foundry tools
foundry_api = FoundryRESTEndpoint()
await foundry_api.register_tool("code_analyzer", analyze_code)
# Mount endpoints
app.include_router(foundry_api.router, prefix="/foundry/v1")
# Use via HTTP
# POST /foundry/v1/code_analyzer
# {
# "code": "def hello(): print('world')",
# "language": "python"
# }10/
├── README.md # This documentation
├── requirements.txt # Python dependencies
├── foundry_tools/
│ ├── __init__.py # Package initialization
│ ├── core/
│ │ ├── __init__.py
│ │ ├── tool_base.py # Base tool implementation
│ │ ├── registry.py # Tool registry
│ │ ├── validation.py # Schema validation
│ │ └── client.py # Foundry Local client
│ ├── integrations/
│ │ ├── __init__.py
│ │ ├── langchain.py # LangChain integration
│ │ ├── semantic_kernel.py # Semantic Kernel integration
│ │ ├── rest_api.py # REST API framework
│ │ ├── cli.py # Command-line interface
│ │ └── jupyter.py # Jupyter notebook tools
│ ├── frameworks/
│ │ ├── __init__.py
│ │ ├── fastapi_tools.py # FastAPI integration
│ │ ├── streamlit_tools.py # Streamlit integration
│ │ ├── gradio_tools.py # Gradio integration
│ │ └── flask_tools.py # Flask integration
│ └── tools/
│ ├── __init__.py
│ ├── code_tools.py # Code analysis tools
│ ├── content_tools.py # Content processing tools
│ ├── data_tools.py # Data analysis tools
│ ├── research_tools.py # Research and retrieval tools
│ └── decision_tools.py # Decision support tools
├── examples/
│ ├── basic_tools.py # Simple tool examples
│ ├── langchain_demo.py # LangChain integration
│ ├── semantic_kernel_demo.py # Semantic Kernel demo
│ ├── rest_api_server.py # REST API server
│ ├── cli_application.py # CLI application
│ ├── jupyter_notebook.ipynb # Interactive notebook
│ ├── streamlit_app.py # Streamlit application
│ └── production_deployment.py # Production patterns
├── integrations/
│ ├── vscode_extension/ # VS Code extension
│ ├── github_actions/ # CI/CD workflows
│ ├── azure_functions/ # Serverless deployment
│ └── docker_containers/ # Containerization
└── tests/
├── test_tools.py # Tool tests
├── test_integrations.py # Integration tests
└── test_frameworks.py # Framework tests
from foundry_tools import FoundryTool
from typing import List, Dict, Any
@FoundryTool(
name="summarize_content",
description="Summarize long-form content into key points",
model="phi-4-mini",
parameters={
"content": {"type": "string", "description": "Content to summarize"},
"max_points": {"type": "integer", "description": "Maximum summary points", "default": 5},
"style": {"type": "string", "description": "Summary style", "enum": ["bullet", "paragraph", "outline"]}
}
)
async def summarize_content(
content: str,
max_points: int = 5,
style: str = "bullet"
) -> Dict[str, Any]:
"""Summarize content using Foundry Local model."""
# The decorator automatically handles:
# - Parameter validation
# - Foundry Local client setup
# - Error handling and logging
# - Response formatting
system_prompt = f"""
Summarize the following content into {max_points} key points.
Use {style} format for the summary.
"""
# This gets automatically routed to Foundry Local
return {
"summary": "Generated summary here...",
"points": max_points,
"style": style,
"word_count": len(content.split())
}from foundry_tools.core import BaseFoundryTool
class CodeAnalysisTool(BaseFoundryTool):
"""Advanced code analysis tool with state management."""
name = "advanced_code_analyzer"
description = "Perform comprehensive code analysis"
model = "phi-4-mini"
def __init__(self):
super().__init__()
self.analysis_cache = {}
self.supported_languages = ["python", "javascript", "typescript", "java", "csharp"]
async def validate_input(self, **kwargs) -> bool:
"""Custom input validation."""
language = kwargs.get("language", "").lower()
return language in self.supported_languages
async def execute(self, code: str, language: str, analysis_type: str = "full") -> Dict[str, Any]:
"""Execute code analysis."""
# Check cache
cache_key = f"{hash(code)}_{language}_{analysis_type}"
if cache_key in self.analysis_cache:
return self.analysis_cache[cache_key]
# Perform analysis using Foundry Local
result = await self.foundry_call({
"system_prompt": f"Analyze this {language} code for {analysis_type} analysis",
"user_prompt": f"Code to analyze:\n\n```{language}\n{code}\n```",
"max_tokens": 1000
})
# Process and cache result
analysis_result = self.process_analysis_result(result, analysis_type)
self.analysis_cache[cache_key] = analysis_result
return analysis_result
def process_analysis_result(self, raw_result: str, analysis_type: str) -> Dict[str, Any]:
"""Process the raw analysis result into structured data."""
# Implementation here
passfrom foundry_tools import StreamingFoundryTool
from typing import AsyncGenerator
@StreamingFoundryTool(
name="code_generator",
description="Generate code with real-time streaming",
model="qwen2.5-coder-0.5b"
)
async def generate_code(
specification: str,
language: str = "python",
include_tests: bool = False
) -> AsyncGenerator[Dict[str, Any], None]:
"""Generate code with streaming responses."""
# Yield metadata first
yield {
"type": "metadata",
"language": language,
"include_tests": include_tests,
"estimated_lines": 50
}
# Stream code generation
async for chunk in foundry_stream({
"prompt": f"Generate {language} code: {specification}",
"stream": True
}):
yield {
"type": "code_chunk",
"content": chunk.content,
"complete": chunk.finish_reason is not None
}
# Yield final result
if include_tests:
async for test_chunk in foundry_stream({
"prompt": f"Generate unit tests for the above {language} code",
"stream": True
}):
yield {
"type": "test_chunk",
"content": test_chunk.content,
"complete": test_chunk.finish_reason is not None
}from langchain.agents import AgentExecutor, create_openai_functions_agent
from langchain.prompts import ChatPromptTemplate
from foundry_tools.langchain import FoundryToolkit
# Create Foundry-powered toolkit
toolkit = FoundryToolkit()
toolkit.add_tool("code_analyzer", model="phi-4-mini")
toolkit.add_tool("content_summarizer", model="qwen2.5-0.5b")
toolkit.add_tool("research_assistant", model="phi-3.5-mini")
# Create agent with Foundry tools
prompt = ChatPromptTemplate.from_messages([
("system", "You are a helpful assistant with access to Foundry Local tools."),
("user", "{input}"),
("assistant", "{agent_scratchpad}")
])
agent = create_openai_functions_agent(
llm=toolkit.get_llm(), # Uses Foundry Local as LLM
tools=toolkit.get_tools(),
prompt=prompt
)
agent_executor = AgentExecutor(agent=agent, tools=toolkit.get_tools())
# Use the agent
result = await agent_executor.ainvoke({
"input": "Analyze this Python code and summarize any issues you find"
})from semantic_kernel import Kernel
from semantic_kernel.connectors.ai.open_ai import OpenAIChatCompletion
from foundry_tools.semantic_kernel import FoundryKernelPlugin
# Initialize kernel with Foundry Local
kernel = Kernel()
# Add Foundry Local as chat service
foundry_service = OpenAIChatCompletion(
service_id="foundry_chat",
ai_model_id="phi-4-mini",
api_key="not-needed",
base_url="http://localhost:5273/v1"
)
kernel.add_service(foundry_service)
# Create and add Foundry plugin
foundry_plugin = FoundryKernelPlugin()
foundry_plugin.add_function("analyze_code", model="phi-4-mini")
foundry_plugin.add_function("summarize_text", model="qwen2.5-0.5b")
kernel.add_plugin(foundry_plugin, plugin_name="foundry_tools")
# Use in Semantic Kernel workflows
result = await kernel.invoke(
"foundry_tools",
"analyze_code",
code="def hello(): print('world')",
language="python"
)from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from foundry_tools.rest import FoundryRESTFramework
app = FastAPI(title="Foundry Local Tools API")
# Initialize Foundry REST framework
foundry_framework = FoundryRESTFramework()
# Auto-register all available tools
await foundry_framework.auto_register_tools([
"code_analyzer",
"content_summarizer",
"data_processor",
"research_assistant"
])
# Mount Foundry endpoints
app.include_router(
foundry_framework.get_router(),
prefix="/api/v1/foundry",
tags=["foundry-tools"]
)
# Custom endpoint using Foundry tools
class AnalysisRequest(BaseModel):
code: str
language: str = "python"
@app.post("/api/v1/analyze")
async def analyze_code_endpoint(request: AnalysisRequest):
try:
result = await foundry_framework.call_tool(
"code_analyzer",
code=request.code,
language=request.language
)
return {"success": True, "analysis": result}
except Exception as e:
raise HTTPException(status_code=500, detail=str(e))
# Health check endpoint
@app.get("/api/v1/health")
async def health_check():
status = await foundry_framework.get_health_status()
return {
"foundry_status": status.foundry_running,
"active_models": status.loaded_models,
"available_tools": status.available_tools
}import typer
from rich.console import Console
from rich.table import Table
from foundry_tools.cli import FoundryCLI
app = typer.Typer(name="foundry-tools")
console = Console()
foundry_cli = FoundryCLI()
@app.command()
async def analyze(
file_path: str = typer.Argument(..., help="Path to code file"),
language: str = typer.Option("python", help="Programming language"),
output: str = typer.Option("table", help="Output format (table, json, yaml)")
):
"""Analyze code file using Foundry Local."""
try:
with open(file_path, 'r') as f:
code = f.read()
result = await foundry_cli.call_tool(
"code_analyzer",
code=code,
language=language
)
if output == "table":
table = Table(title=f"Code Analysis: {file_path}")
table.add_column("Metric", style="cyan")
table.add_column("Value", style="magenta")
for key, value in result.items():
table.add_row(key, str(value))
console.print(table)
elif output == "json":
console.print_json(data=result)
else:
console.print(result)
except Exception as e:
console.print(f"[red]Error: {e}[/red]")
raise typer.Exit(1)
@app.command()
async def list_tools():
"""List all available Foundry tools."""
tools = await foundry_cli.list_available_tools()
table = Table(title="Available Foundry Tools")
table.add_column("Name", style="cyan")
table.add_column("Description", style="white")
table.add_column("Model", style="yellow")
for tool in tools:
table.add_row(
tool["name"],
tool["description"][:50] + "..." if len(tool["description"]) > 50 else tool["description"],
tool["model"]
)
console.print(table)
if __name__ == "__main__":
app()from foundry_tools import CompositeFoundryTool
@CompositeFoundryTool(
name="full_code_review",
description="Comprehensive code review using multiple analysis tools"
)
async def comprehensive_code_review(code: str, language: str = "python") -> Dict[str, Any]:
"""Perform comprehensive code review using multiple tools."""
# Run multiple analyses in parallel
analyses = await asyncio.gather(
call_tool("code_analyzer", code=code, language=language),
call_tool("security_scanner", code=code, language=language),
call_tool("performance_analyzer", code=code, language=language),
call_tool("style_checker", code=code, language=language)
)
# Synthesize results
return await call_tool("analysis_synthesizer", analyses=analyses)from foundry_tools.context import ContextAwareFoundryTool
class ProjectAnalyzerTool(ContextAwareFoundryTool):
"""Analyze entire project with context awareness."""
async def execute(self, project_path: str, analysis_depth: str = "shallow") -> Dict[str, Any]:
"""Analyze project with full context."""
# Build project context
context = await self.build_project_context(project_path)
# Analyze with context
return await self.foundry_call_with_context({
"prompt": f"Analyze this {context.language} project",
"context": context.to_dict(),
"analysis_depth": analysis_depth
})
async def build_project_context(self, project_path: str) -> ProjectContext:
"""Build comprehensive project context."""
# Implementation here
passfrom foundry_tools.chains import FoundryToolChain
# Define a tool chain for document processing
doc_processing_chain = FoundryToolChain([
("extract_text", {"input": "document_path"}),
("summarize_content", {"input": "extracted_text", "style": "outline"}),
("generate_keywords", {"input": "summary"}),
("create_metadata", {"input": ["summary", "keywords"]})
])
# Execute the chain
result = await doc_processing_chain.execute({
"document_path": "/path/to/document.pdf"
})from foundry_tools.cache import CacheConfig, CacheStrategy
cache_config = CacheConfig(
strategy=CacheStrategy.LRU,
max_size=1000,
ttl=3600, # 1 hour
key_generator="content_hash"
)
# Apply to specific tools
@FoundryTool(
name="cached_analyzer",
cache_config=cache_config
)
async def cached_code_analyzer(code: str) -> Dict[str, Any]:
# Expensive analysis that benefits from caching
passfrom foundry_tools.pool import ModelPoolConfig
pool_config = ModelPoolConfig(
models={
"phi-4-mini": {"instances": 2, "priority": "high"},
"qwen2.5-coder-0.5b": {"instances": 1, "priority": "medium"},
"phi-3.5-mini": {"instances": 1, "priority": "low"}
},
load_balancing="round_robin",
health_check_interval=30
)
# Configure tool registry with pool
registry = FoundryToolRegistry(model_pool_config=pool_config)from foundry_tools.batch import BatchProcessor
@BatchProcessor(
batch_size=10,
timeout=60,
parallel_batches=3
)
async def batch_code_analysis(code_files: List[str]) -> List[Dict[str, Any]]:
"""Process multiple code files in batches."""
results = []
for code_file in code_files:
with open(code_file, 'r') as f:
code = f.read()
result = await call_tool("code_analyzer", code=code)
results.append(result)
return resultsfrom foundry_tools.monitoring import ToolMetrics
# Automatic metrics collection
metrics = await ToolMetrics.get_tool_performance("code_analyzer")
print(f"Average execution time: {metrics.avg_execution_time}s")
print(f"Success rate: {metrics.success_rate}%")
print(f"Cache hit rate: {metrics.cache_hit_rate}%")from foundry_tools.health import HealthMonitor
health_monitor = HealthMonitor()
# Monitor tool health
health_status = await health_monitor.check_all_tools()
print(f"Healthy tools: {health_status.healthy_count}")
print(f"Failed tools: {health_status.failed_tools}")from foundry_tools.analytics import UsageAnalytics
analytics = UsageAnalytics()
# Track tool usage patterns
usage_report = await analytics.generate_usage_report(
start_date="2024-01-01",
end_date="2024-01-31"
)
print(f"Most used tool: {usage_report.most_used_tool}")
print(f"Peak usage time: {usage_report.peak_usage_time}")इस उदाहरण को पूरा करने के बाद, आप समझेंगे:
-
टूल इंटीग्रेशन पैटर्न
- फंक्शन-आधारित और क्लास-आधारित टूल डिज़ाइन
- Microsoft Foundry Local इंटीग्रेशन पैटर्न
- स्कीमा सत्यापन और टाइप सुरक्षा
- त्रुटि प्रबंधन और पुनर्प्राप्ति
-
फ्रेमवर्क इंटीग्रेशन
- LangChain टूल विकास
- Semantic Kernel फंक्शन इंटीग्रेशन
- REST API फ्रेमवर्क इंटीग्रेशन
- CLI एप्लिकेशन विकास
-
उत्पादन विचार
- प्रदर्शन अनुकूलन रणनीतियाँ
- कैशिंग और संसाधन प्रबंधन
- निगरानी और अवलोकन
- सुरक्षा और सत्यापन
-
उन्नत टूल पैटर्न
- टूल कंपोज़िशन और चेनिंग
- संदर्भ-संवेदनशील प्रसंस्करण
- बैच और स्ट्रीमिंग संचालन
- कस्टम इंटीग्रेशन विकास
- इंटीग्रेशन प्रोजेक्ट्स: अपने पसंदीदा फ्रेमवर्क्स के साथ कस्टम इंटीग्रेशन बनाएं
- टूल विकास: अपने डोमेन के लिए विशेष टूल्स बनाएं
- प्रदर्शन ट्यूनिंग: अपने विशिष्ट उपयोग मामलों के लिए अनुकूलित करें
- उत्पादन परिनियोजन: एंटरप्राइज़ उपयोग के लिए टूल्स को स्केल करें
योगदान निर्देशों के लिए मुख्य रिपॉजिटरी दिशानिर्देश देखें।
यह उदाहरण Microsoft Foundry Local प्रोजेक्ट के समान लाइसेंस का पालन करता है।