Chatting with Agents
The chat interface is the primary way you interact with agents. You ask questions in natural language and the agent queries your connected data sources to build an answer.
Starting a Conversation
Section titled “Starting a Conversation”You can open a chat with an agent in two ways:
- From the agent card — Click on any agent in the Agents portal to open its chat.
- From the sidebar — If you have pinned an agent to the sidebar, click it there for quick access.
Asking Questions
Section titled “Asking Questions”Type your question in the input field and send it. The agent interprets your question, determines which connected data sources to query, and executes the necessary tool calls to gather the information it needs.
You do not need to specify which data source to use. The agent decides automatically based on the question and the connections available to it. For example, asking “Why are we seeing elevated error rates on the API?” will prompt a Debugger agent to check your log provider, and potentially your code repository, without any manual direction.
How Agents Reason
Section titled “How Agents Reason”Each response can involve up to 15 reasoning steps. During these steps, the agent may:
- Query one or more connected data sources
- Analyze and correlate the returned data
- Make follow-up queries based on initial findings
- Synthesize a final answer from all gathered evidence
This multi-step reasoning allows agents to handle questions that require information from several sources or that need iterative investigation.
Web Search
Section titled “Web Search”Agents can also search the web and fetch URL content to supplement their answers. The specific web capabilities depend on the AI provider configured for the agent:
| Provider | Capabilities |
|---|---|
| Anthropic | Web search and URL fetching |
| Google Search and URL context | |
| OpenAI | Web search |
Web search is useful when the agent needs external context — for example, checking whether a third-party service is experiencing an outage, or looking up documentation for a library.
Response Formatting
Section titled “Response Formatting”Agent responses are rendered as markdown. This includes:
- Formatted text with headings and lists
- Code blocks with syntax highlighting
- Tables for structured data
- Inline code references
Conversation History
Section titled “Conversation History”Message history is persisted automatically. When you return to an agent’s chat, your previous conversation is still there, so you can continue where you left off or ask follow-up questions that build on earlier answers.
Message Limits
Section titled “Message Limits”Chat messages count toward your monthly tier allocation:
| Plan | Messages per Month |
|---|---|
| Free | 100 |
| Pro | Unlimited |
| Business | Unlimited |
When you reach the Free tier limit, you will need to wait until the next billing cycle or upgrade your plan to continue chatting.