Chat And Timeline

Chat is where you talk to an agent. The timeline is the broader work record: messages, plans, tool activity, generated files, approvals, pending requests, and completed deliverables.

Open an agent timeline

1. Open Agents.
2. Select an agent.
3. Open Chat.

If you are in an organization, confirm the correct personal or organization context before assuming an agent is missing.

What appears in the timeline

• Messages from you, collaborators, approved contacts, app channels, or the agent.
• Plan updates when the platform breaks a request into steps.
• Deliverables such as summaries, files, tables, briefs, or reports.
• Tool activity showing meaningful work, external actions, searches, file operations, or app usage.
• Files and attachments uploaded by people or created by the platform.
• Pending requests for approval, credentials, contact access, human input, or another agent.
• Processing state when the agent is still working.

The timeline is the best place to diagnose whether the agent is waiting, working, blocked, or finished.

Send a useful message

Good messages include:

• The desired outcome.
• Important constraints.
• The expected output format.
• Files or sources to use.
• What should require approval.

Example:

Use the uploaded renewal spreadsheet to identify the five highest-risk vendors. Create a short table with risk, reason, owner, and recommended next step. Do not email anyone yet.

Optional follow-up suggestions

After a reply, the timeline may offer one useful next step. It offers none when the request is complete or no useful next step is clear. Selecting a suggestion sends it as a normal message.

Suggestions clear when you send a new message or request a stop. After a stop, they stay hidden until you send another message and the agent replies. Stopping the current response does not permanently cancel every related task; explain the new direction or explicitly ask to resume when appropriate.

Use Hide → Hide for now to dismiss the current set. Use Hide → Turn off suggestions to hide future suggestions too. You can re-enable them in chat Settings.

Work with files in chat

Attach files directly in the composer when they are part of the current request. Use the agent's filespace when files should be reused across future work.

Typical flow:

1. Add one or more files in the composer.
2. Include the instruction for how the agent should use them.
3. Send the message.
4. Review any file references or generated outputs in the timeline.

For durable file management, use Files And Workspaces.

Respond to pending requests

When the agent needs a human decision, the timeline may show a pending request. Common examples:

• A question with options or free text.
• A request for a credential or environment-style secret.
• A request to contact a person.
• A request to create, connect, brief, or change another agent.

Read the request carefully. Approve only the exact action you intend. If the request is too broad, reply with a narrower instruction.

See Approvals And Requests.

Stop or redirect work

If an agent is moving in the wrong direction:

• Send a corrective message if it can safely continue.
• Use stop controls when available if the current work should halt.
• Narrow the next instruction before restarting.
• Review inference Usage and pause the agent if it is consuming more than intended.

Collaborate in chat

Collaborators can work with the same agent when they have access. Permissions and visible controls depend on ownership, organization context, and collaborator role.

For access management, see Collaborators And Ownership.

Practical habits

• Use one primary request per message when the work is complex.
• Put durable instructions in the charter, not only in chat.
• Use filespace paths or filenames when referring to reusable files.
• Ask the platform to summarize its plan before external communication.
• Keep credentials out of chat. Use Secrets And Credentials.

Related workflows

Use Broadcast To Agents for a shared instruction across multiple chats, Schedules And Reminders for timed work, and Agent Lifecycle to distinguish stopping a run from pausing an agent. If work is stuck, use Troubleshooting.