Conversations

Conversations are how you interact with your agents. Each conversation is an independent session where you can ask questions, give instructions, and have the agent perform tasks.

Starting a conversation

1. From the Agents page — Click on an agent, then navigate to the Conversations tab and click New Conversation.
2. From the Conversations page — Click Conversations in the sidebar, select an agent, and start chatting.

Screenshot: The conversation interface showing the chat input, streamed agent response, and tool call indicators.

The chat interface

Sending messages

Type your message in the input field and press Enter. The agent streams its response in real time.

You can attach files to your messages. Supported file types depend on the agent's configuration and connected tools.

Agent reasoning

When the Think Tool is enabled, you may see a collapsible reasoning section before the agent's response. This shows the agent's internal thought process.

Tool calls

When the agent uses a connected tool, a tool call indicator shows:

• The tool being used (e.g., "Sending email via Gmail")
• The parameters being passed
• The result returned by the tool

Approval requests

If the agent proposes an action that requires approval, the conversation pauses and displays Approve and Deny buttons. Review the details before deciding.

Managing conversations

Conversation history

Conversations are automatically saved when Save Messages is enabled on the agent. Access previous conversations from the Conversations page, listed in reverse chronological order.

Session compaction

For long conversations, the agent summarizes earlier messages into a compact summary, keeping context manageable while preserving important details.

Interrupting the agent

If the agent is mid-response or in a multi-step tool chain, you can interrupt it to cancel the current operation and redirect.

Tips for effective prompting

• Be specific — Instead of "help me with email," try "draft a follow-up email to the client about the project timeline."
• Provide context — The more relevant context you give, the better the response.
• Use follow-ups — Build on previous messages rather than starting fresh.
• Reference documents — Mention topics you want the agent to search for in connected knowledge bases.