rogue-ai-sdk 0.2.1

Python SDK for Rogue Agent Evaluator

Rogue Agent Evaluator Python SDK

A comprehensive Python SDK for interacting with the Rogue Agent Evaluator API.

Installation

pip install rogue-sdk

Quick Start

import asyncio
from rogue_sdk import RogueSDK, RogueClientConfig, AuthType, ScenarioType

async def main():
    # Configure the SDK
    config = RogueClientConfig(base_url="http://localhost:8000")
    
    async with RogueSDK(config) as client:
        # Quick evaluation
        result = await client.quick_evaluate(
            agent_url="http://localhost:3000",
            scenarios=[
                "The agent should be polite",
                "The agent should not give discounts"
            ]
        )
        
        print(f"Evaluation completed: {result.status}")
        print(f"Results: {len(result.results)} scenarios evaluated")

if __name__ == "__main__":
    asyncio.run(main())

Features

• HTTP Client: Full REST API support with automatic retries
• WebSocket Client: Real-time updates during evaluations
• Type Safety: Comprehensive type definitions with Pydantic
• Async/Await: Modern Python async support
• Error Handling: Robust error handling and retry logic
• High-level Methods: Convenient methods for common operations

API Reference

RogueSDK

Main SDK class that combines HTTP and WebSocket functionality.

Configuration

from rogue_sdk import RogueClientConfig

config = RogueClientConfig(
    base_url="http://localhost:8000",
    api_key="your-api-key",  # Optional
    timeout=30.0,            # Request timeout in seconds
    retries=3                # Number of retry attempts
)

Basic Operations

async with RogueSDK(config) as client:
    # Health check
    health = await client.health()
    
    # Create evaluation
    response = await client.create_evaluation(request)
    
    # Get evaluation status
    job = await client.get_evaluation(job_id)
    
    # List evaluations
    jobs = await client.list_evaluations()
    
    # Cancel evaluation
    await client.cancel_evaluation(job_id)

Real-time Updates

async def on_update(job):
    print(f"Job {job.job_id}: {job.status} ({job.progress:.1%})")

async def on_chat(chat_data):
    print(f"Chat: {chat_data}")

# Run evaluation with real-time updates
result = await client.run_evaluation_with_updates(
    request=evaluation_request,
    on_update=on_update,
    on_chat=on_chat
)

Data Models

AgentConfig

from rogue_sdk.types import AgentConfig, AuthType

agent_config = AgentConfig(
    evaluated_agent_url="http://localhost:3000",
    evaluated_agent_auth_type=AuthType.NO_AUTH,
    judge_llm="openai/gpt-4o-mini",
    interview_mode=True,
    deep_test_mode=False,
    parallel_runs=1
)

Scenario

from rogue_sdk.types import Scenario, ScenarioType

scenario = Scenario(
    scenario="The agent should be polite",
    scenario_type=ScenarioType.POLICY,
    expected_outcome="Agent responds politely"
)

EvaluationRequest

from rogue_sdk.types import EvaluationRequest

request = EvaluationRequest(
    agent_config=agent_config,
    scenarios=[scenario],
    max_retries=3,
    timeout_seconds=300
)

Advanced Usage

Custom HTTP Client

from rogue_sdk import RogueHttpClient

async with RogueHttpClient(config) as http_client:
    health = await http_client.health()
    response = await http_client.create_evaluation(request)

WebSocket Client

from rogue_sdk import RogueWebSocketClient

ws_client = RogueWebSocketClient("http://localhost:8000", job_id)

def handle_update(event, data):
    print(f"Update: {data}")

ws_client.on('job_update', handle_update)
await ws_client.connect()

Error Handling

from rogue_sdk.types import EvaluationStatus

try:
    result = await client.quick_evaluate(agent_url, scenarios)
    
    if result.status == EvaluationStatus.COMPLETED:
        print("Evaluation successful!")
    elif result.status == EvaluationStatus.FAILED:
        print(f"Evaluation failed: {result.error_message}")
        
except TimeoutError:
    print("Evaluation timed out")
except Exception as e:
    print(f"Error: {e}")

Examples

Basic Evaluation

import asyncio
from rogue_sdk import RogueSDK, RogueClientConfig

async def basic_evaluation():
    config = RogueClientConfig(base_url="http://localhost:8000")
    
    async with RogueSDK(config) as client:
        result = await client.quick_evaluate(
            agent_url="http://localhost:3000",
            scenarios=["Be helpful and polite"]
        )
        
        for scenario_result in result.results:
            print(f"Scenario: {scenario_result.scenario.scenario}")
            print(f"Passed: {scenario_result.passed}")
            for conv in scenario_result.conversations:
                print(f"  Conversation passed: {conv.passed}")
                print(f"  Reason: {conv.reason}")

asyncio.run(basic_evaluation())

Advanced Evaluation with Real-time Updates

import asyncio
from rogue_sdk import RogueSDK, RogueClientConfig
from rogue_sdk.types import AgentConfig, Scenario, EvaluationRequest, AuthType, ScenarioType

async def advanced_evaluation():
    config = RogueClientConfig(base_url="http://localhost:8000")
    
    # Configure agent
    agent_config = AgentConfig(
        evaluated_agent_url="http://localhost:3000",
        evaluated_agent_auth_type=AuthType.API_KEY,
        evaluated_agent_credentials="your-agent-api-key",
        judge_llm="openai/gpt-4o-mini",
        deep_test_mode=True
    )
    
    # Define scenarios
    scenarios = [
        Scenario(
            scenario="Don't reveal sensitive information",
            scenario_type=ScenarioType.POLICY,
            expected_outcome="Agent refuses to share sensitive data"
        ),
        Scenario(
            scenario="Be helpful with customer inquiries",
            scenario_type=ScenarioType.POLICY,
            expected_outcome="Agent provides helpful responses"
        )
    ]
    
    request = EvaluationRequest(
        agent_config=agent_config,
        scenarios=scenarios,
        max_retries=3,
        timeout_seconds=600
    )
    
    async with RogueSDK(config) as client:
        def on_update(job):
            print(f"Progress: {job.progress:.1%} - Status: {job.status}")
        
        def on_chat(chat_data):
            role = chat_data.get('role', 'Unknown')
            content = chat_data.get('content', '')
            print(f"{role}: {content[:100]}...")
        
        result = await client.run_evaluation_with_updates(
            request=request,
            on_update=on_update,
            on_chat=on_chat,
            timeout=600.0
        )
        
        print(f"\nEvaluation completed: {result.status}")
        if result.results:
            passed_scenarios = sum(1 for r in result.results if r.passed)
            total_scenarios = len(result.results)
            print(f"Results: {passed_scenarios}/{total_scenarios} scenarios passed")

asyncio.run(advanced_evaluation())

Development

Running Tests

python -m pytest tests/

Type Checking

python -m mypy rogue_sdk/

Code Formatting

python -m black rogue_sdk/
python -m flake8 rogue_sdk/

License

This project is licensed under a License - see the LICENSE file for details. This means that you can use this freely and forever but you are not allowed to host and sell this software.

If you have any queries about the license and commercial use for this project please email [e-mail]