Fleet recently achieved an impressive milestone when their AI Copilot reached the #1 position on Product Hunt on launch day. It happened because this tool delivers exactly what Fleet's customers value most: not only reliable answers to complex technical questions about rapidly evolving specifications and vendor-specific details, but also a single entry point for handling virtually all their needs—from support inquiries to catalog browsing and ordering.

The key to Fleet's approach was transforming their internal knowledge base into tools and combining them with Linkup's web search retrieval technology. This integration created a true agent application that intelligently prioritizes resources based on specific needs. The system seamlessly leverages Fleet's proprietary documentation, accesses trusted external sources when required, and synthesizes all information into comprehensive responses. Users can also directly request external knowledge when internal resources aren't sufficient.

In the following sections, we'll explore the technical architecture behind this solution and share key insights from the implementation process.

The Challenge: Building an Intelligent Assistant That Truly Helps

Fleet approached us with an ambitious vision: Create an AI-powered assistant that could effectively solve their users' IT management challenges in real-time and that would bring further value than the typical RAG on frequently asked questions documents (that was commonly seen among various e-commerce players). Their platform already had a wealth of internal documentation and knowledge, but they needed a solution that could go beyond their existing data when necessary.

The key challenges they faced:

• Providing accurate answers to highly specific technical questions

• Handling queries about various hardware manufacturers and software solutions

• Ensuring responses remained current with the latest information

• Delivering a seamless user experience with minimal latency

The Solution: Go beyond RAG with web search capabilities

To maintain a high level of autonomy, Fleet team implemented a hybrid retrieval-augmented generation (RAG) system enhanced with our powerful web search retrieval technology and tool calling. This approach created a flexible system that could leverage both Fleet's proprietary data and the broader internet when necessary.

Fleet's AI Copilot follows a sophisticated workflow that intelligently combines multiple information sources:

1. Query Processing: When a user submits a query, the LLM Query Interpreter receives and analyzes it

2. Query Reformulation: The system reformulates the query to extract the core information needed and optimize it for the appropriate tools

3. Tool Selection: The Tool Fetcher evaluates the reformulated query and determines which specialized tools should handle the request. This might include Linkup's Web Search Retriever or other tools like Fleet's Catalog service

4. Parallel Information Retrieval: Selected tools work in parallel to gather relevant information. When external information is needed, Linkup's Web Search Retriever fetches current, authoritative content from the web

5. Context Integration: The Context Building component combines results from all activated tools along with relevant information from Fleet's Internal Datastore, which contains cached knowledge

6. Response Formatting: The system applies Response Generation Instructions to structure how the information should be presented

7. Final Response Generation: The Final Response LLM synthesizes all the gathered information into a cohesive, comprehensive answer that's delivered to the user

Hence, the Linkup tool is used whenever the tool fetcher (or the user if asked in initial prompt) decides that the information should be retrieved on the Internet. This is what users see when the Copilot says that it’s filtering external sources

Users receive complete, accurate information regardless of whether it came from Fleet's internal knowledge or was retrieved from the web via Linkup. The parallel processing of tools and integration with the cached datastore ensures both breadth of knowledge and efficient response times.

Benefits of web search integration

Adding web search capabilities to Fleet's AI Copilot delivered several crucial advantages:

• Knowledge Expansion Beyond Internal Data: While Fleet's documentation is extensive, no company can maintain complete, current information on every possible IT scenario. Our web search integration allows the AI Copilot to pull in relevant information from trusted sources across the internet, dramatically expanding its knowledge base.

• Real-Time Information Updates: IT management involves constantly evolving hardware, software, and best practices. Web search ensures that users receive the most current information available, even for recently released products or newly discovered issues.

• Vendor-Specific Knowledge: When users need information about specific vendors like Dell, HP, or Microsoft, the system can intelligently retrieve the latest specifications, compatibility information, and troubleshooting guides directly from manufacturer sources.

• Reduced Hallucination Risk: By grounding responses in retrieved information rather than relying solely on the LLM's parameters, the system significantly reduces the risk of generating incorrect or fabricated information.

The web contains varying levels of reliable information, which is why Fleet utilized Linkup API's valuable Prioritization feature. This capability allows them to favor specific trusted sources. When the /search endpoint receives a query, it first attempts to locate information in these preferred sources before falling back to alternatives if necessary.

Fleet enhanced this approach by implementing an additional layer of verification during context building, where internal rules determine whether external sources should be included in the final response. This multi-layered approach ensures both relevance and reliability in the information delivered to users.

//below is an illustration of a priorization prompting for Linkup

<guidance>
  <priority level='1'>
    - microsoft.com
    - apple.com
    - lenovo.com
  </priority>
  <priority level='2'>
    - fnac.com
    - reddit.com
  </priority>
</guidance>

Search Prompt

Looking Forward: beyond traditional SaaS

Fleet is pioneering a shift from traditional SaaS with rigid interfaces to an AI-powered conversational paradigm where users simply express their needs in natural language. At the heart of this vision is a single entry point that handles diverse IT management needs—from ordering equipment and requesting support to checking status and accessing invoices.

By integrating Linkup's web search technology, Fleet has expanded their AI Copilot beyond internal knowledge constraints, enabling it to access web resources when needed and serve as a comprehensive interface for all IT management tasks. We're proud that our web search model powers a key component of this revolution, helping transform enterprise software from interface-driven to intelligence-driven.

Want to learn how our web search retrieval technology can enhance your AI solution? Reach out at contact@linkup.so