Anyone who has handed a bigger job to an AI agent knows how the work tends to unfold.
The user describes what they want, the agent starts planning and carrying out the steps, and for a while everything moves along. Then the smaller questions appear. The user might wonder why the agent picked one method instead of another, or need to check a piece of data it is relying on, or have an idea surface that they would like to test without changing the main direction.
Most of the time those thoughts go straight into the same chat.
The original goal stays there, but so do all the side remarks.
After enough of them the thread grows long, the main purpose becomes harder to see at a glance, and the user spends extra time scrolling back or restating details just to keep the agent on track.
That pattern is common when the task itself takes more than a few minutes. Building a small application, pulling together research from several sources, or assembling a report all tend to produce moments where the user wants a quick answer without stopping the larger effort. In ordinary chat windows those moments usually force a choice: either interrupt the main work or open an entirely new session and lose the context already built.
Neither option feels ideal when the agent is already partway through a multi-step job.
Manus addressed this by adding a feature called 'Side Chat.'
While the agent is still working on the main task, users can open a separate panel next to it.
The new panel already knows what the main task has done so far, so they can ask a question, get an answer, and close it again without touching the primary conversation. The main thread stays focused on the original goal, and the side questions stay out of the way.
In practice it means users do not have to pause the agent or start a completely new session every time something small comes up. Users can leave the long-running job alone and handle the quick clarifications beside it.
If more than one question appears, users can open additional side chats rather than stacking everything into one place. Each side chat remains light and temporary, meant for those moments when information is needed but the direction of the main work should stay the same.
The idea fits the way many people already use agents for bigger jobs such as building an app, gathering research, or putting together a report.
The agent keeps making progress on its own, while the user dips in and out with the smaller questions that naturally arise. The result is a cleaner record of the main effort and less time spent managing the conversation itself.
It is distinct from the earlier Branch feature, which splits a conversation into separate parallel sessions for different outputs. Side Chat is specifically designed as a lightweight, context-aware side panel for ad-hoc questions alongside an active long-running task.



























































































































































































































































































































































































