Perplexity's AI Agent search browser, 'Comet' is already in beta But how does "Agentic search" work? let's understand together... If you checked my last AI Agent updates, Perplexity has started rolling out Comet for beta testing for selective Mac users. Now the details about Comet are not available but the term "Agentic Search" is something we already know. If you look closely, you see it is somewhat similar to Agentic RAG. To fact-check my theory, I referred to a research blog written by a Global Lead Architect at Google, Glean AI's architecture, and the Agentic RAG research paper. π Let's understand it together: 1. User Query: The process begins with a user submitting a query. 2. Agentic Search Console: The query is received by the Agentic Search Console, which acts as the central hub for processing the request. Here, the query is first parsed and then concatenated with a system prompt for an efficient response from LMs. - Agentic Reasoning: 3. Language Models- The modified input is processed by a large language model or Large Reasoning model, which then sends an instruction to the retrieval agent. - (Switch): The modern agentic search engines now have an LM switch. With a press of a button, you could switch between models like GPT-4o or o1. 4. Retrieval Agents: To gather searched information, the system employs Retrieval Agents, which connects to external tools and knowledge sources: 5. Retrieval Agent A: This agent pulls data from a knowledge base, including a Vector Database and Semantic Database, to retrieve relevant information stored in different formats. 5. Retrieval Agent B: Simultaneously, This agent links with search engines (like Google, Bing, X, or LinkedIn) to fetch real-time or external data from the web and social platforms. 6. Compiled Output: The insights and retrieved responses are then sent to the Retrieval Agent to understand and consolidate the responses. 7. LMs: Depending Upon the model, they will understand the context of the input and the retrieved information to build up their reasoning to personalize the answer according to the prompt. 8. Memory: Throughout the reasoning process, the system accesses both short-term and long-term memory to maintain context, recall previous interactions, or store new information relevant to the query. 9. Output to User: Finally, the compiled output is delivered back to the user as the response to their original query, completing the workflow. If you are a business leader, we've developed frameworks that cut through the hype, including our five-level Agentic AI Progression Framework to evaluate any agent's capabilities in my latest book. π Book info: https://amzn.to/4irx6nI Save πΎ β React π β Share β»οΈ & follow for everything related to AI Agents | 22 comments on LinkedIn
