Artisan IMG > API Operation (api-operation-trigger) (b7cd35d2-96c1-4ed0-a03c-d8a2a7d0d4eb)
Artisan SVG > Loop Collection (loop) (synchronize-1)
Artisan SVG > Data Storage (storage) (server-edit-1)

Knowledge Agent

Project
Artificial Intelligence
Intermediate

This is a 'Project' template which means that it contains a group of workflows that work together to achieve a particular aim

Overview
Copy

This accelerator enables you to create an AI-powered knowledge agent that searches across multiple sources to provide comprehensive answers to your team's questions. The agent combines knowledge from webscraping with web search capabilities to deliver accurate, contextualized responses.

Key Features
Copy

  • Public Documentation Search: Leverage your organization's public documentation

  • Google Search Integration: Augment responses with current web information

  • Intelligent Response Generation: Uses advanced LLMs to synthesize information from multiple sources

Prerequisites
Copy

  1. Before implementing this template, you will need:

    1. To make sure your organization has this AI feature enabled (Admin access required)

    2. Access to a Tray.io instance with API credentials

    3. URLs for your public documentation

    4. Authentication token for Google Search API

Getting Live
Copy

Initial Setup
Copy

  1. Create a new project

  2. Select "Merlin Agent Builder"

  3. Select the Knowledge Agent accelerator

  4. Create or select required authentications

Configure Knowledge Sources
Copy

  1. Scrape your Public Documentation

    1. Add documentation URLs to the scraping form or the manual scraping workflow

    2. Set up crawling depth

  2. Google Search Integration

    1. Add your Google Search API credentials

Deploy the Agent
Copy

  1. Enable the main workflow

  2. Test with sample queries

  3. Monitor performance and adjust the system prompt as needed

Key Workflow Components
Copy

Knowledge Processing
Copy

  • Documentation Crawler: Indexes public documentation

  • Vector Creation: Generates embeddings for all content

  • Knowledge Storage: Manages vector database operations

Query Processing
Copy

  • Query Understanding: Agent reasons over the user questions

  • Response Generation: Synthesizes information into coherent answers

  • Session Management: Maintains context across invocations of the agent that are related

Search Integration
Copy

  • Search tool: Agent determines when to use Google Search

  • Source Attribution: Tracks and cites information sources

Best Practices
Copy

  1. Regularly update your knowledge base

  2. Monitor and adjust RAG thresholds (both in the search tool and knowledge tool)

  3. Review and refine response quality

  4. Implement feedback loops for continuous improvement

  5. Add more tools that have access to internal knowledge (CRM, Internal Documentation, etc)

  6. Run evaluations - you can follow our guidance here