fencio-proxy-client 0.1.0

Python SDK for Fencio HTTP Proxy - Agent observability and tracing

Fencio Proxy Client SDK

Python SDK for integrating with the Fencio HTTP proxy for agent observability and tracing.

Installation

pip install -e .

Or with optional dependencies:

# For requests library support
pip install -e ".[requests]"

# For httpx library support
pip install -e ".[httpx]"

# For all HTTP libraries
pip install -e ".[all]"

Quick Start

from fencio_proxy_client import FencioProxyClient
import requests

# Initialize the client
client = FencioProxyClient(
    agent_id="my_threat_agent",
    api_key="your_api_key_here",
    proxy_url="http://localhost:8080"
)

# Start routing requests through Fencio proxy
client.start()

# All HTTP requests are now traced
response = requests.get("http://httpbin.org/get")

# Stop when done
client.stop()

Context Manager Support

from fencio_proxy_client import FencioProxyClient
import requests

# Use as a context manager
with FencioProxyClient(agent_id="my_agent", api_key="secret") as client:
    # Requests inside this block are traced
    response = requests.get("http://httpbin.org/get")

# Automatically stops when exiting the block

Layer Classification

You can classify requests by layer (e.g., LLM calls vs tool calls):

from fencio_proxy_client import FencioProxyClient
import requests

client = FencioProxyClient(agent_id="my_agent", api_key="secret")
client.start()

# Mark LLM requests
with client.layer_context('llm', {'model': 'gpt-4', 'provider': 'openai'}):
    response = requests.post(
        'https://api.openai.com/v1/chat/completions',
        json={'model': 'gpt-4', 'messages': [{'role': 'user', 'content': 'Hello'}]}
    )

# Mark tool requests
with client.layer_context('tool', {'tool': 'web_search'}):
    response = requests.get('http://api.example.com/search?q=test')

client.stop()

Session Management

client = FencioProxyClient(
    agent_id="my_agent",
    api_key="secret",
    session_id="custom_session_123"  # Optional: specify your own session ID
)

# Or let the SDK generate a session ID automatically
client = FencioProxyClient(
    agent_id="my_agent",
    api_key="secret",
    auto_session=True  # Default: generates UUID session ID
)

# Generate a new session ID on the fly
new_session = client.new_session()
print(f"Started new session: {new_session}")

Supported HTTP Libraries

The SDK automatically patches these HTTP libraries:

• requests - Most popular HTTP library
• httpx - Modern async-capable HTTP client
• urllib3 - Low-level HTTP library

You can choose which libraries to patch:

# Only patch requests
client = FencioProxyClient(
    agent_id="my_agent",
    api_key="secret",
    patch_libraries=['requests']
)

# Patch multiple libraries
client = FencioProxyClient(
    agent_id="my_agent",
    api_key="secret",
    patch_libraries=['requests', 'httpx']
)

Integration with LangChain Agents

from langchain.agents import initialize_agent, Tool
from langchain.llms import OpenAI
from fencio_proxy_client import FencioProxyClient

# Initialize Fencio client
fencio = FencioProxyClient(
    agent_id="langchain_threat_agent",
    api_key="your_api_key"
)
fencio.start()

# Define tools
tools = [
    Tool(
        name="ThreatIntel",
        func=lambda x: requests.get(f"http://threat-api.com/lookup?q={x}").json(),
        description="Look up threat intelligence data"
    )
]

# All LLM calls and tool calls are now traced through Fencio
llm = OpenAI(temperature=0)
agent = initialize_agent(tools, llm, agent="zero-shot-react-description")
result = agent.run("What is the threat level for IP 192.168.1.1?")

fencio.stop()

Configuration

Environment Variables

You can also configure the client using environment variables:

export FENCIO_AGENT_ID="my_agent"
export FENCIO_API_KEY="secret_key"
export FENCIO_PROXY_URL="http://localhost:8080"

import os
from fencio_proxy_client import FencioProxyClient

client = FencioProxyClient(
    agent_id=os.getenv('FENCIO_AGENT_ID'),
    api_key=os.getenv('FENCIO_API_KEY'),
    proxy_url=os.getenv('FENCIO_PROXY_URL', 'http://localhost:8080')
)

API Reference

FencioProxyClient

__init__(agent_id, api_key=None, proxy_url="http://localhost:8080", session_id=None, auto_session=True, patch_libraries=None)

Initialize the Fencio proxy client.

Parameters:

• agent_id (str): Unique identifier for this agent
• api_key (str, optional): API key for authentication
• proxy_url (str): URL of the Fencio proxy server (default: "http://localhost:8080")
• session_id (str, optional): Session identifier (auto-generated if not provided)
• auto_session (bool): Auto-generate session ID if not provided (default: True)
• patch_libraries (list, optional): List of HTTP libraries to patch (default: all available)

start()

Start routing HTTP requests through Fencio proxy.

stop()

Stop routing HTTP requests through Fencio proxy.

is_active() -> bool

Check if the client is currently active.

get_headers(layer=None, layer_metadata=None) -> dict

Get Fencio headers to inject into HTTP requests.

Parameters:

• layer (str, optional): Layer classification (e.g., 'llm', 'tool')
• layer_metadata (dict, optional): Additional metadata for the layer

Returns:

• Dictionary of HTTP headers

new_session() -> str

Generate a new session ID and update the client.

Returns:

• The new session ID

layer_context(layer, metadata=None)

Context manager for setting layer classification for a block of code.

Parameters:

• layer (str): Layer classification (e.g., 'llm', 'tool')
• metadata (dict, optional): Additional metadata for the layer

How It Works

1. Library Patching: When you call client.start(), the SDK patches common HTTP libraries (requests, httpx, urllib3) to route all HTTP traffic through the Fencio proxy.

2. Header Injection: For each HTTP request, the SDK automatically injects these headers:

   • X-Fencio-Agent-ID: Your agent identifier
   • X-Fencio-API-Key: Your API key (if provided)
   • X-Fencio-Session-ID: Session identifier
   • X-Fencio-Layer: Layer classification (if using layer_context)
   • X-Fencio-Layer-Metadata: JSON-encoded layer metadata

3. Proxy Routing: All HTTP/HTTPS requests are automatically routed through the Fencio proxy at the specified proxy_url.

4. Trace Storage: The Fencio proxy captures complete request/response data and stores it with agent context for later analysis.

License

MIT License