Skip to content

Workspace

The Workspace at /workspace is where most people spend their time in ExoHuman. It is a focused chat app — separate from the operator screens — where you hold ongoing conversations with your agents, answer what they need from you and see what they produced.

Everyone with access to an agent can use the Workspace, including people with the user role and external clients an agent was shared with by email.

  • Sidebar (left): your Inbox, projects, New chat, search, your agents roster, starred chats, chats grouped by date and archived chats.
  • Conversation (centre): the chat with the selected agent, with tabs for each open chat.
  • Rail (right): context about the agent and this conversation.

Your principal assistant, Your assistant, is pinned at the top of the sidebar. See Principal assistant.

Every person–agent pair has a pinned Main chat: the place the agent reaches you when nothing else names a chat — a scheduled brief, a question it raises, a message it starts. Open more chats for separate topics:

  • New chat in the header, or ⌘J / Ctrl+J, starts a fresh chat with the current agent.
  • Mention another agent with @ to open a room with both; see Agent network.
  • Close a chat with its × — it is archived, never deleted, and can be reopened.
  • Reset on Main archives the current Main and starts the agent fresh.

A Workspace chat resumes the same underlying agent session on every turn, so the agent keeps its tool results, intermediate state and reasoning between messages, not just the text. Long conversations are compacted automatically when they approach the model’s context limit (around 85%), keeping a summary of earlier work.

  • Enter sends, Shift+Enter adds a line. Messages can be up to 8,000 characters.
  • Type / for the agent’s skills, or @ to bring in another agent or tag a person.
  • Attach files by clicking the paperclip, dropping them on the conversation or pasting them.
  • Dictation turns speech into text; the call button starts a voice conversation.
  • A model selector lets you choose a capability tier for agents that support it.
  • Send becomes Stop while the agent works; stopping returns your words to the composer.

Unsent text is kept as a draft per chat in your browser.

Replies stream live. A card under your message shows the agent’s status, elapsed time and what it is doing right now (Reading…, Running…, Delegating to…), with Stop and Open in Work. Closing the browser doesn’t stop the work — the reply is waiting when you come back.

Under every reply you can copy it, reply to that specific message, and rate it Helpful or Not helpful. A negative rating notifies the agent’s owner in Operations with your optional comment.

TabShows
InfoThe agent’s role, suggestions for you, and recorded decisions
CanvasThe agent’s live canvases
FilesFiles you sent and files the agent shared
ProjectsProjects this agent works on
AsksThe agent’s questions and approvals waiting on you
WorkExecutions from this chat and the agent’s history
LoopsRun and watch loops

On the Info tab, Decisions records what you and the agent approved, deferred or rejected, why, which alternatives were considered, and when to review it. Active decisions are read into every turn, so the agent reuses the reasoning next time. Decisions can be reconfirmed, corrected, reversed or closed — never silently deleted.

KeysAction
⌘J / Ctrl+JNew chat with the current agent
⌥↓ / ⌥↑Next / previous agent
⌘. / Ctrl+.Show or hide the rail
⌘/ / Ctrl+/Search agents and chats
⌥/Show all shortcuts
EscStop the running turn

Share an agent with someone’s email and they can sign in to the Workspace with a 6-digit code — no password and no platform account. They see only the agents shared with them and the canvases those agents publish to their audience. Client sessions renew while used and end after a period of inactivity.