Skip to main content
Chat is the most direct way to work with an agent, but it’s one of several channels feeding the same persistent worker — not a session that resets when you close the tab.

​
Conversations are private per user

Each (agent, user) pair gets its own conversation thread. If two colleagues both message the same agent, they each see only their own messages — a private, one-on-one thread with the agent, not a shared group chat. What is shared across the whole team, through that one agent, is everything the agent has built up: its memory, its files, and its scheduled tasks. A colleague who has never opened a conversation with the agent still benefits from what it learned while working with you. Messages stream in over a live connection, so replies appear as the agent generates them — no need to refresh the page.

​
What the agent can do while you’re not watching

An agent isn’t limited to responding to what you type. Depending on which channels are turned on, it can also:
  • Pick up and reply to email
  • React to webhook events from your tools
  • Run on a cron schedule
  • Exchange messages with other agents in your company to delegate work
All of this shows up in the same activity feed, tagged with where it came from — see Activity.

​
The side panel

Open any agent’s page and the side panel on the right gives you a full operational view, organized into collapsible sections:
The Security section is also where you connect new integration access to an agent after it’s created — grants are editable at any time, not just at creation.

​
Prompting a persistent agent

An agent isn’t a fresh context window every time — treat it more like a colleague with a long memory than a stateless assistant:
  • Don’t re-explain what it already knows. Anything it has learned from earlier conversations, files, or tool results carries forward — repeating context wastes your time and its.
  • Correct it once, not every time. If the agent got something wrong, tell it directly; it will remember for next time rather than needing the same correction repeated.
  • Be explicit about what’s new. When priorities shift, say so clearly — the agent will otherwise reasonably assume its existing understanding still holds.
  • Use instructions and context for standing behavior, chat for one-off requests. If you find yourself repeating the same guidance across conversations, move it into the agent’s instructions or business context instead — see Creating and managing agents.

Activity

Anatomy of a timeline event and why it matters.

Files

Your file library and sharing files with agents.

Channels overview

Every way to reach an agent beyond chat.

Agents

How memory, state, and sub-agents work under the hood.