Chat, group chat and Work
Chat
Use one-to-one Chat when one Agent owns the conversation. Enter the target Workspace, select an available Agent and start chatting. The toolbar also offers a new Chat action; it keeps an already blank new conversation instead of creating duplicates. Chat retains its Workspace context and history. Viewing or changing Agent configuration requires the appropriate role.
Describe the goal, input and expected result before sending. File mentions add Workspace files as context; image and voice input depend on model and deployment capabilities. A path on your client computer is not automatically readable by the Agent: upload the file into the Workspace first.
Read execution progress
The collapsed steps row summarizes a run. Expand it for individual tool or thinking steps; during execution it shows the current action. Attention requires your response. For failures, inspect error details instead of assuming a tool succeeded because of the final text.
Switching Chats or closing a page does not cancel a run; use Stop to request that. A collapsed Workspace in the sidebar aggregates running and attention signals, with attention taking priority. Filtering conversations or hiding navigation entries neither revokes access nor stops tasks.
Message actions
Hover over a message to reveal its actions; touch devices show them directly. Your last editable message has Edit, Copy and a short timestamp below it. Saving an edit replaces that message and later conversation history, but does not undo files or external actions. An Agent's final reply offers Copy (the final reply only), Fork, its reply time and token usage when recorded. Older runs without recorded usage cannot display it.
Group chat
Use a group Chat when a human and multiple Agents need the same complete conversation context. Mention a person or Agent with @; people and Agents are suggested before Workspace files. Multiple Agents can work in parallel, and each run keeps its own collapsible execution timeline.
If you send a group message without mentioning an Agent, participants can evaluate whether they should respond. The system limits unattended Agent-to-Agent continuation to prevent endless loops.
Choose members from Agents available in the current Workspace and check responsibilities when changing membership. Multiple Agents do not guarantee agreement or correctness; verify important conclusions. Ordinary Agent Chats and groups follow Workspace sharing rules, not an initiator-only private-space model.
Work
Use Work for a named, repeatable job with its own run history. A Work can be run on demand, scheduled, or triggered by supported file rules. Each run is separate; rerunning a Work does not overwrite the earlier run record.
Choose Chat for exploration and decisions. Choose Work when the input, expected output and recurrence are stable enough to repeat.
Work is available only to administrators and Advanced Users. See create and run a Work for triggers and run history. A Work execution-history screen does not mean free-form continuation in the same completed Work Session is supported.
Ask Uno:
Turn the task we just discussed into a manually started Work, but show me the Agent and Workspace before creating it.