NNyquest docs

Quick Start

Get your first agent run in under a minute.

1. Click the Agent Mode button

In the left icon-nav rail (between Settings and Admin), click the 🤖 Agent Mode button.

The chat surface morphs into the agent panel: a goal input, a max-steps selector, and a Run button.

2. Type a goal

Be specific. The clearer the goal, the better the agent's tool selection.

Good examples:

  • "What is the current price of GLD?"
  • "Find the latest Tesla earnings news, then look up TSLA's price"
  • "Summarize the top 3 results for 'Meta AI Reality Labs spending 2025'"

Vague examples (will work but give weaker answers):

  • "Find something interesting"
  • "Tell me about cars"

3. (Optional) Adjust the max steps

The default is 6 steps per run. Most goals finish in 1–3. If you have a research-heavy goal that needs more iterations, raise it. Maximum is 20 (absolute ceiling).

4. Click Run agent

(Or press Cmd/Ctrl+Enter.)

The agent starts thinking. You'll see step cards stream in real time.

5. Watch the timeline

Each card shows:

  • Step number (1, 2, 3...)
  • Tool emoji (📈 quote_lookup, 🔎 web_search, etc.)
  • Tool name
  • Running cost in cents
  • Expand chevron — click to see the planner's reasoning, the args passed, and a result summary

6. Read the final answer

When the agent finishes, the final answer renders as a markdown card with the full response — same rendering as a regular chat reply. Tables, code blocks, lists, headings all work.

The agent's answer is also appended to your regular chat history with the label "Agent Mode · N steps", and the UI auto-flips back to chat mode so you can continue the conversation normally.

7. Continue the conversation

Once the answer lands in chat history, ask follow-ups like you would in normal chat. The regular chat completion has the agent's answer in context, so you can say "can you elaborate on point 3?" or "what does that mean for me?"

To run another agent goal, click the Agent Mode button again.

What if I want to stop a run mid-flight?

Click Stop in the AgentTimeline footer. The agent will finish whatever step it's currently on, then synthesize an answer from what it has. See Stopping a Run for details.