Skip to main content
A Circuit agent has access to your workspace’s documents. It can search, read, and reason over that knowledge to answer your questions. The key to getting good results is understanding how to communicate with it.

Think colleague, not search engine

The most important shift is to stop thinking in keywords and start thinking in chat. A Circuit agent understands natural language, context, and nuance.
pump specs XYZ series
This gives the agent very little to work with. It doesn’t know what kind of specs you need, why you need them, or what you’ll do with the answer.

Provide context

The more context you give, the better the answer. Consider including:
  • What you’re working on: “I’m preparing an RFP response” or “I’m troubleshooting a unit in the field”
  • What you already know: “The customer is using the 2024 model” or “I’ve already checked the installation guide”
  • What format you need: “Give me a summary I can paste into an email” or “List the steps in order”
  • Any constraints: “The customer requires metric units” or “I need information from the latest revision only”

Have a chat

You don’t have to get the perfect question on the first try. Circuit agents support multi-turn chats, so you can refine as you go.
1

Ask your initial question

Start with what you know, even if it’s broad.
What are the maintenance intervals for the ABC compressor line?
2

Narrow down based on the response

If the answer covers too much or not enough, tell the agent.
I’m specifically looking at the ABC-500 model. What’s the recommended schedule for the air filter and oil changes?
3

Ask for a different format if needed

Can you put that in a table with the component, interval, and procedure?

When the answer isn’t right

If the agent gives you an incorrect or unhelpful response, don’t just re-ask the same question. Instead, explain what was wrong:
  • “That’s the old model. I need info on the current version.”
  • “The answer is too general. Can you give me the specific part numbers?”
  • “That document is outdated. Do you have anything from 2024 or later?”
The agent learns from the chat context and adjusts its approach.

Next steps