English

Português

English

Português

Appearance

title: Testing Agents on SipPulse AI description: Learn how to effectively test agents using the SipPulse AI platform, including best practices, testing methodology, and performance optimization techniques. ​

SipPulse AI - Agent Testing Documentation ​

1. Opening a Chat with an Agent ​

The SipPulse AI platform allows you to test and interact with various conversational agents. This section covers how to access and initiate a chat session with an agent.

Accessing the Agents Dashboard ​

  1. Navigate to the SipPulse AI agent management interface
  2. The main dashboard displays all available agents with the following information:
    • NAME: The agent's identifier and internal reference code
    • DESCRIPTION: Brief explanation of the agent's purpose
    • MODEL: The AI model powering the agent (e.g., gpt-4o-mini, gemini-2.0-flash)
    • TOOLS: Available actions for each agent

Starting a Chat Session ​

To open a chat with any agent:

  1. Locate the desired agent in the list (e.g., "Receptionist [DEMO DO NOT REMOVE]")
  2. Click the chat icon in the TOOLS column or select "Open chat" from the dropdown menu
  3. A new chat interface will open, displaying:
    • The agent's name and identification
    • A conversation window where messages will appear
    • Connection details (UI, Room ID)
    • Model information and conversation metrics

Conversation Threading System ​

Each interaction with an agent is organized into a unique thread, regardless of the communication channel used:

  1. Thread Identification:

    • Each conversation receives a unique thread identifier (e.g., "thr_0196ce1d5d2a74df85d8b0c51f1")
    • This identifier appears in the interface and is used for internal tracking

  2. Multichannel Monitoring:

    • The system allows you to track all conversations regardless of the medium used
    • Threads can be initiated and continued through different channels:
      • WebRTC (for browser-based video/audio calls)
      • SIP API (for integration with telephony systems)
      • Playground (for testing via web interface)

  3. Unified History:

    • All interactions are recorded in the same thread, creating a continuous history
    • In the interface, you can view previous conversations by date and time
    • Messages display indicators of the channel used (e.g., "UI User")

  4. Conversation Management:

    • You can filter and search for specific conversations using their thread identifiers
    • Closed conversations can be reopened by clicking on the indicated option

2. Importance of Extensive Testing for Voice Agents ​

One of the most critical stages in developing conversational agent solutions is conducting comprehensive testing before deployment. This practice is essential for the following reasons:

Performance and Reliability Assessment ​

  • Verification of conversation flows: Ensures the agent responds correctly to various inputs and scenarios
  • Measurement of correct execution percentage: Allows quantification of agent reliability in real situations
  • Identification of failure points: Helps discover and fix problems before contact with real customers

Cost-Benefit Analysis ​

  • Cost per interaction: Tests allow calculation of the average cost of each service
  • Prompt optimization: Enables adjustments to reduce token consumption while maintaining quality
  • Financial planning: Provides data to estimate operational costs at scale

Efficient Troubleshooting ​

  • Thread history as an investigative tool: The threading system provides complete traceability of interactions, being fundamental for:
    • Identifying error patterns or unexpected behaviors
    • Analyzing specific problematic conversations
    • Comparing different versions of the same agent
    • Documenting use cases for future reference
    • Facilitating continuous refinement of agents based on real interactions

This process of iterative testing and analysis is essential to ensure that SipPulse AI voice agents provide reliable, efficient, and cost-controlled experiences when implemented in production environments.

3. Agent Functionality Examples ​

As shown in the image, the Digital Receptionist agent demonstrates key functionalities:

  • Company introduction: "Good morning! You've reached V.Office, a Cisco and Microsoft partner specializing in networking and information security since 1997."
  • Call transfers: When a caller asks to speak with a specific person, the agent can execute a transfer_call() function with the appropriate SIP address
  • Natural responses: "Certainly! Please hold while I transfer you to Ana Cristina."
  • Performance metrics: The interface displays token usage (38.3 tokens/s), response time, and cost per interaction (R$ 0.03)

These capabilities make the digital receptionist an effective front-end for business communications, handling initial greetings and routing calls appropriately.