Recorded as an episode of AI Engineer Spotlight, this session opens up the three components every Tars agent is made of: the agent that holds the goal and the prompt, the knowledge it can retrieve from, and the tools it can act with. The demo agent is trained on a live product documentation site.

The technical half is the reason to watch. Knowledge is not a separate concept bolted onto the agent, it is packaged as a retrieval tool the agent chooses to call, and the session shows both what happens to your content before an agent ever sees it and the retrieval settings that are usually hidden. Watch this if you are evaluating Tars technically and want to see the mechanics rather than a polished demo.

What this video covers

• The three parts of an agent, the agent itself holding the goal and prompt, the knowledge connected as a retrieval tool, and the tools that let it act
• A question answered from live product documentation, with the agent choosing to call retrieval and running a semantic search behind the answer
• How an interested visitor's details get captured into a connected sheet, with the tool calls visible in the thread and the note that the same slot could hold a CRM instead
• What the knowledge module does to your content before indexing, cleaning it, splitting it, embedding it, and storing it for semantic search
• The retrieval settings that usually stay hidden, how many chunks to pull, how many to pass to the model, and whether hybrid search is on
• How a short plain-English instruction generates both the prompt and the welcome message the agent opens with

Chapters

• 0:00 Introductions and what the platform is
• 0:40 The three parts, agent, knowledge, and tools
• 1:20 A question answered from live documentation
• 2:50 Capturing a visitor's details into a connected sheet
• 4:40 How knowledge is packaged as a retrieval tool
• 5:23 What the knowledge module does to your content
• 7:11 Generating a prompt from a short instruction