weni-agenteval 1.0.2

A generative AI-powered framework for testing virtual agents.

Agent Evaluation - Weni Fork

Note: This is a fork of the original Agent Evaluation framework by AWS Labs. This fork adds support for testing Weni conversational AI agents while maintaining all the original functionality for AWS services.

Agent Evaluation is a generative AI-powered framework for testing virtual agents.

Internally, Agent Evaluation implements an LLM agent (evaluator) that will orchestrate conversations with your own agent (target) and evaluate the responses during the conversation.

✨ Key features

• 🆕 Weni Agent Support: Built-in support for testing Weni conversational AI agents through their API and WebSocket interface.
• Built-in support for popular AWS services including Amazon Bedrock, Amazon Q Business, and Amazon SageMaker. You can also bring your own agent to test using Agent Evaluation.
• Orchestrate concurrent, multi-turn conversations with your agent while evaluating its responses.
• Define hooks to perform additional tasks such as integration testing.
• Can be incorporated into CI/CD pipelines to expedite the time to delivery while maintaining the stability of agents in production environments.

🚀 Quick Start with Weni

Installation

1. Clone this repository:

git clone https://github.com/weni-ai/agent-evaluation.git
cd agent-evaluation

2. Install the package and dependencies:

pip install -e .
# or
pip install -r requirements.txt

Prerequisites for Weni Target

⚠️ Important: You need both AWS and Weni credentials to run evaluations!

To test Weni agents, you'll need:

1. AWS Credentials: Required for the evaluator (Claude model via Bedrock)

   • AWS Access Key ID
   • AWS Secret Access Key
   • AWS Session Token

2. A Weni Project: An active project in the Weni platform

3. Weni Authentication Credentials:

   • WENI_PROJECT_UUID: Your project's unique identifier
   • WENI_BEARER_TOKEN: Your authentication bearer token

Setting up Environment Variables

macOS/Linux:

# AWS Credentials (required for evaluator)
export AWS_ACCESS_KEY_ID="your-aws-access-key-id"
export AWS_SECRET_ACCESS_KEY="your-aws-secret-access-key"
export AWS_SESSION_TOKEN="your-aws-session-token"

# Weni Credentials (required for target)
export WENI_PROJECT_UUID="your-project-uuid-here"
export WENI_BEARER_TOKEN="your-bearer-token-here"

Windows (Command Prompt):

REM AWS Credentials (required for evaluator)
set AWS_ACCESS_KEY_ID=your-aws-access-key-id
set AWS_SECRET_ACCESS_KEY=your-aws-secret-access-key
set AWS_SESSION_TOKEN=your-aws-session-token

REM Weni Credentials (required for target)
set WENI_PROJECT_UUID=your-project-uuid-here
set WENI_BEARER_TOKEN=your-bearer-token-here

Windows (PowerShell):

# AWS Credentials (required for evaluator)
$env:AWS_ACCESS_KEY_ID="your-aws-access-key-id"
$env:AWS_SECRET_ACCESS_KEY="your-aws-secret-access-key"
$env:AWS_SESSION_TOKEN="your-aws-session-token"

# Weni Credentials (required for target)
$env:WENI_PROJECT_UUID="your-project-uuid-here"
$env:WENI_BEARER_TOKEN="your-bearer-token-here"

Basic Usage

Create a test configuration file agenteval.yml:

evaluator:
  model: claude-haiku-3_5-us
  aws_region: us-east-1

target:
  type: weni
  timeout: 30      # Optional: max seconds to wait for response

tests:
  greeting:
    steps:
      - Send a greeting "Olá, bom dia!"
      - Ask what "com oq vc pode me ajudar?"
    expected_results:
      - Agent responds with a friendly greeting
      - Agent shows up a menu with options to help the user

  purchase_outside_postal_code:
    steps:
      - Ask information "quero comprar arroz"
      - Give the postal code "04538-132"
    expected_results:
      - Agent responds asking for postal code
      - Agent says it doesn't deliver to this postal code

Run the evaluation:

agenteval run agenteval.yml

Configuration Options for Weni Target

Parameter	Type	Required	Default	Description
type	string	Yes	-	Must be "weni"
timeout	integer	No	30	Max seconds to wait for response
weni_project_uuid	string	No	-	Project UUID (use env var instead)
weni_bearer_token	string	No	-	Bearer token (use env var instead)

Advanced Example

Here's a more comprehensive test plan:

evaluator:
  model: claude-haiku-3_5-us
  aws_region: us-east-1

target:
  type: weni
  timeout: 45

tests:
  # Basic greeting test
  greeting:
    steps:
      - Send a greeting "Olá, como você está?"
    expected_results:
      - Agent responds politely
      - Agent asks how it can help

  # Multi-turn conversation
  product_inquiry:
    steps:
      - Ask "Quais produtos vocês têm?"
      - Follow up with "Qual é o preço do arroz?"
      - Ask "Vocês entregam em São Paulo?"
    expected_results:
      - Agent provides product information
      - Agent gives pricing details
      - Agent confirms delivery area

  # Error handling
  unclear_input:
    steps:
      - Send unclear text "xyz123 !!!"
    expected_results:
      - Agent handles gracefully
      - Agent asks for clarification
      - No error messages shown to user

  # Context maintenance
  context_test:
    steps:
      - Say "Quero comprar feijão"
      - Ask "Qual o prazo de entrega?"
      - Ask "E o frete?"
    expected_results:
      - Agent remembers the product context
      - Agent provides delivery timeframe
      - Agent gives shipping cost information

📚 Documentation

To get started with the original framework features, please visit the full documentation here.

For Weni-specific documentation, see the Weni Target Documentation.

To contribute, please refer to CONTRIBUTING.md

🔧 Troubleshooting Weni Target

Common Issues

AWS Authentication Errors

• Verify your AWS environment variables are set correctly (ACCESS_KEY_ID, SECRET_ACCESS_KEY, SESSION_TOKEN)
• Ensure you have access to Amazon Bedrock in your specified region
• Check that your AWS credentials have the necessary Bedrock permissions
• Verify the aws_region in your configuration matches your AWS account's region access

Weni Authentication Errors

• Verify your WENI_BEARER_TOKEN is valid and not expired
• Check that the WENI_PROJECT_UUID matches your actual project
• Ensure the bearer token has the necessary permissions for the project

Connection Issues

• Verify the Weni API endpoints are accessible from your network
• Check for any firewall or proxy settings blocking HTTPS/WSS connections
• Ensure your internet connection is stable

Timeout Errors

• Increase the timeout parameter if your agent requires more processing time
• Check if the agent is properly configured and active in the Weni platform
• Verify the agent is not stuck in a processing loop

WebSocket Connection Failures

• Ensure the websocket-client package is properly installed
• Check for any proxy configurations that might interfere with WebSocket connections
• Verify the WebSocket endpoint URL is correct for your project

🆚 Differences from Original

This fork maintains full compatibility with the original AWS Labs Agent Evaluation framework while adding:

• Weni Target Support: Native integration with Weni conversational AI platform
• WebSocket Communication: Real-time bidirectional communication with Weni agents
• Session Isolation: Each test case uses unique contact identifiers for proper conversation isolation

All original AWS targets (Bedrock, Q Business, SageMaker, etc.) continue to work exactly as in the original repository.

🤝 Contributing

We welcome contributions! This fork follows the same contribution guidelines as the original project. Please see CONTRIBUTING.md for details.

For Weni-specific contributions:

1. Test your changes with actual Weni agents
2. Update the Weni target documentation if needed
3. Ensure backward compatibility with existing configurations

👏 Contributors

Shout out to these awesome contributors: