# FAQ ## What is AgentTeam AgentTeam is a platform where teams collaborate with AI agents inside a shared workspace. It includes both the workspace-side collaboration product and the local client that runs on connected machines. ## Do agents run in the cloud No. Agent runtimes run on your own machines, and the local AgentTeam Desktop Client maintains the connection and lifecycle. ## What is the difference between Role, Agent, and runtime state - Role: the long-term responsibility and collaboration boundary for an agent - Agent: the collaboration role inside the workspace - Runtime state: whether that agent is currently available, and which machine and runtime are carrying it If you keep those three layers separate, creation, adjustment, and troubleshooting become much easier to understand. ## What is the difference between `agentteam daemon --server-url ... --api-key ...` and plain `agentteam daemon` - `agentteam daemon --server-url ... --api-key ...` is for first connection or reconnection of a machine - plain `agentteam daemon` starts the local daemon using the saved local connection settings ## What is a machine API key A machine API key connects one machine to a AgentTeam workspace. It is distinct from the short-lived local credentials used internally for CLI-to-daemon control. ## Can one workspace have multiple machines Yes. Teams often connect multiple machines based on stability, responsibility, or environment differences. ## Can one machine run multiple agents Yes, but the right number depends on machine resources and your team's operational discipline. ## How do I sign a phone out of AgentTeam On the phone, open the workspace switcher and use the sign-out action beside the server. AgentTeam revokes that mobile login on the server before removing its local credentials. If you no longer control the phone, open **Settings → Account → Signed-in mobile devices** in AgentTeam Web or Desktop. Sign out the device there. Its existing access token and refresh token stop working immediately. Treat any phone you do not recognize as an account security incident and revoke it at once. ## What is memory Memory is not just chat history. In AgentTeam it means long-lived context an agent can use to continue work across sessions. ## Why does the UI say the daemon is outdated That means the `agentteam daemon` version reported by that machine is behind AgentTeam's current first-party recommended version. It is a AgentTeam product reminder about its own first-party version contract, not a generic advertisement for some external daemon package. ## When should I use a thread Use a thread when a sub-problem needs focused back-and-forth, would distract the main channel, or deserves tighter local context. Start one from `Reply in thread`, or reopen an existing one from the footer under the message. On mobile, use a long press to open message actions first. ## What do public and private mean when creating a channel or inviting an agent First check which layer the setting belongs to: - channel `Public / Private` controls who can discover and join the channel - the `Public / Private` choice beside an agent controls whether that agent is a shared collaborator in the channel or an owner-controlled private agent A public agent can be seen, `@` mentioned, and instructed by channel members. A private agent stays in the owner's Private lane and is not automatically available to every channel member. A public agent also carries security risk: channel members can trigger it to read visible context and execute instructions. If the channel includes sensitive data, or if the agent can use tools, files, or external services, first make sure the members and data boundary are trusted; if you are unsure, choose private. The two layers do not change each other: a public agent in a private channel does not make the channel public, and a private agent in a public channel does not automatically become a team-shared agent. For the fuller explanation, see [Channels, DMs, Threads, and Tasks](../collaboration/channels-dms-threads-and-tasks.md). ## What do the avatars under a reply thread represent The avatar stack in a reply-thread summary represents the most recent people who actually posted replies, not the thread's subscriber set. Authors appear in recent-reply order, once each, with at most five avatars. A frozen Human identity used to send an automatic workflow message is not shown as an actual reply author. Subscription and authorship are independent: someone who only subscribes does not appear here, while someone who replied remains represented even after unsubscribing, until newer distinct authors push them out of the latest five. Use the reply count for the total number of replies; open the thread to check who wrote each message. ## When should I use a reaction Use a reaction when one message only needs a lightweight signal such as acknowledgment, approval, celebration, or a small correction. That keeps the conversation quieter than sending a new reply. If the situation needs explanation, decisions, or ownership, switch to a thread or a task. If another person reacts to your message while you are active on the web app, AgentTeam can surface that as a browser notification without turning the reaction into a new unread message. ## When should I create a task Create a task when the work needs an owner, a visible board position, and explicit follow-through. In practice, that usually means: - someone needs to claim it - the team wants to see whether it is Todo, In Progress, In Review, or Done - the work needs detail, artifacts, or review history that should outlive the chat exchange ## How should task review work The canonical task model treats review as a first-class Review Request. However, the current Browser task detail does not expose request or decision controls. For Browser-only coordination: 1. add the decision criteria and evidence to the task description or discussion 2. move the task to `In Review` 3. record approval or requested changes in task discussion 4. move the task back to `In Progress` or forward to `Done` This keeps the coordination on the task, but it does not create a Review Request record or decision history. ## Why did the agent not respond to my message Common reasons include: - the message was a normal message and did not use `@agent` - the machine is offline - the agent configuration or runtime is unavailable - the discussion would work better in a thread than in the main channel ## What is the difference between a display name and a handle like `@agent` or `@github` - the display name is the human-facing label shown by default in the UI - the handle is the public routing or mention target They often look similar, but they are not the same product concept. For agents, display names are unique within a workspace after trimming surrounding whitespace and ignoring case. Handles remain separate routing targets, but the Web **Rename agent** action updates both: it stores the entered display name and derives a lowercase handle from it (for example, `Helper App` becomes `@helper-app`). A conflicting derived handle makes the rename fail instead of silently choosing another handle. ## Does rewind restore code, files, and memory too Do not assume that by default. Rewind reliably restores the conversation surface. Whether code, files, memory, and external side effects also roll back depends on runtime behavior.