Appearance

Quickstart ​

Get started with AgentSight in seconds. With just a few lines of code, you can start monitoring your AI agent's conversations, questions, answers, and more. No setup or infrastructure required.

AgentSight is designed to be lightweight and developer-friendly. Instantiate the client with your API key, and you're ready to track your data.

Installation ​

First, install the AgentSight SDK. We also recommend installing python-dotenv to manage your API key securely via environment variables.

bash
pip install agentsight python-dotenv

Initial setup ​

  1. Add your API key to a .env file.
  2. Initialize the tracker in your code using the key.
python
# Load API key and initialize AgentSight
import os
from agentsight import ConversationTracker
from dotenv import load_dotenv

load_dotenv()  # Load environment variables from .env file

# Retrieve the API key
api_key = os.getenv("AGENTSIGHT_API_KEY")

# Initialize the conversation tracker
tracker = ConversationTracker(api_key=api_key)

Setting Your AgentSight API Key ​

You need an AgentSight API key to send data to your AgentSight dashboard. Get your API key from the AgentSight Dashboard. For security and convenience, we recommend setting it as an environment variable.

bash
export AGENTSIGHT_API_KEY="your_agentsight_api_key_here"

Note: If you're using a .env file, make sure to call load_dotenv() before initializing ConversationTracker

This code will automatically use the API key from the .env:

python
from dotenv import load_dotenv
load_dotenv()

tracker = ConversationTracker()

Conversation Tracking ​

Conversations are uniquely identified to help you maintain context across multiple interactions. Once you create a conversation with your ID, use that same ID to reference your tracking methods with that conversation.

Here is how to specify a conversation and conversation_id:

python
conversation_id="your_conversation_id"
tracker.get_or_create_conversation(
    conversation_id=conversation_id
)

Conversation ID Priority

You can only pass conversation ID in get_or_create_conversation. If the conversation_id is not set, we will create one automatically for you which could result in creating new conversation in each iteration.

You can use generate_conversation_id function from agentsight.helpers import generate_conversation_id

How to pass tracking data about the conversation like geographical data, source and other, visit Tracking conversation

Tracking Data ​

Track Question ​

Capture only the user's question without an immediate answer.

python
tracker.track_question("How does machine learning work?")

Read more Track Question

Track Answer ​

Log an AI response independently of a question.

python
tracker.track_answer("Machine learning involves training algorithms...")

Read more Track Answer

Track Token Usage ​

Track usage tokens for your workflows.

python
tracker.track_token_usage(
    prompt_tokens=100,
    completion_tokens=25,
    total_tokens=125
)

Read more Track Token Usage

Track Action ​

Record specific actions or tool usage during an interaction.

python
tracker.track_action(
    action_name="data_retrieval",
    duration_ms=250,
    tools_used={"database": "PostgreSQL"},
    response="Data successfully retrieved"
)

Read more Track Action

Track Button ​

Log user interface interactions and button clicks.

python
tracker.track_button(
    button_event="user_selection",
    label="Confirm Order",
    value="order_id_123"
)

Read more Track Button

Track Attachments ​

Send file attachments separately or with other tracking methods.

python
tracker.track_attachments(
    attachments=[{
        'filename': 'report.pdf',
        'content': base64_encoded_content,
        'mime_type': 'application/pdf'
    }],
    mode='base64' # default
)

Attachment Modes:

  • Base64: Encode files directly in payload
  • Form Data: Send files as multipart form data

Read more Track Attachment

Next Steps ​

Now that you’ve set up the basics, let’s continue with looking into Core Concepts