evoke-sdk 0.2.0

AI Detection & Response SDK - OpenTelemetry-based observability for LLM applications

AI Detection & Response Python-based SDK for monitoring agent workflows, language model execution, tool calls, and relevant data sources.

Features

• Automatic tracing: Works with any language model provider (OpenAI, Anthropic, custom)
• Zero code changes: Just initialize and go
• Comprehensive telemetry: Tracks tools, models, prompts, reasoning
• Simple API: 2 lines of code

Supported Frameworks

• OpenAI SDK
• Anthropic SDK
• Google (Gemini) SDK
• AWS Bedrock SDK
• Azure Foundry SDK
• LiteLLM SDK
• LangChain

Installation

pip install evoke-sdk

Quick Start

import evoke

# Initialize once at startup
evoke.init(api_key="evoke_pk_your_api_key")

# All LLM calls are now automatically traced!
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello!"}]
)

What Gets Tracked

• Models: Which models were used (gpt-4, claude-3, etc.)
• Prompts & Responses: Full input and output data
• Tools: Which tools/functions were called
• Data Sources: Which vector stores/retrievers were accessed
• Tokens: Input/output/cached token counts
• Timing: Duration and timestamps
• Errors: Exceptions and stack traces

Optional: Explicit Tracing

Use the @guard decorator for high-level workflows:

@evoke.guard(name="customer_support_agent")
def handle_customer_query(query: str):
    # Your agent logic
    return agent.run(query)

Optional: Add Custom Context

Link custom business values to track custom events and create custom detections:

@evoke.guard(name="process_order")
def process_order(order_id: str, user_id: str):
    evoke.add_context(
        user_id=user_id,
        order_id=order_id,
        environment="production"
    )
    return process()

Flush Before Exit

Ensure all data is sent before your app exits:

# At the end of your script
evoke.flush()