Talk to an agent

A conversation is the primary way to use an agent. You send a message, the agent works through it, and the result streams back as it's produced.

What you can do

• Stream responses live — watch the agent's reasoning, the tools it uses (such as retriever_search or generate_image), and the final answer appear as it works, including any media it produces.
• Keep context across turns — each conversation is a session with memory, so follow-ups work naturally and you can resume later; sessions are titled automatically.
• Adjust the reasoning — on supported models, set how much the agent thinks before answering, and choose whether to show that thinking.
• Add context to a message — attach documents or collections for the agent to work from, and turn web access on or off.

Example ask

"What changed in this contract since last week?"

What's coming

Suggested follow-up questions, response feedback, and injecting extra context mid-conversation. See the Roadmap.