An agent has a role, a map and a work area
Agent Center supports project workflows and a combined company work area. Use Work areas → Company for one overview across all or selected projects and processes, or Project groups for shared settings with separate execution in each project. The agent’s description explains its purpose; its action map defines the actual sequence of reads, decisions and actions. Connected apps make their granted MCP tools usable in authorized project workflows.
Choose an existing agent or create one for the responsibility you need: client follow-up, delivery coordination, financial review, research or content preparation. A role is a starting point; the configured actions determine what the agent can execute.
Select an agent connected to the project to inspect its role and state. One agent can be used in different project contexts with the relevant project data and permissions.
Build the working route
- 1Choose the agent and work area
Open Agent Center from the application menu. Connect the agent to a project, use Project groups for several separate project workflows, or open Work areas → Company and select a saved agent for one combined company review.
- 2Set the start condition
In the action map choose a manual request, a schedule or an available trigger. Configure the timing and time zone when shown.
- 3Add the first source
Read project tasks, memory, work products or a resource from a connected app. Pick the actual source the later steps will need.
- 4Choose the next action
Use the action chooser on the map. Configure the action and bind its inputs to a preceding output, a selected resource or a fixed value.
- 5Add a decision when needed
Choose the condition and connect its routes, including Otherwise. Define what happens when no task matches or required client data is absent.
- 6Save, check and start
Resolve the concrete configuration issues, save the map and start the workflow in the selected work area. Review the next scheduled run and the outcome of a controlled manual run.
Examples of useful routes
| Responsibility | Source and condition | Result |
|---|---|---|
| Client intake | Read tasks with the client tag; check whether client facts are present | Save available client data to CRM; finish without a write when there is no data. |
| Delivery coordination | Read project tasks and check deadlines or blockers | Prepare a status update or use an available task/calendar action. |
| Research and content | Read a brief, project memory and authorized external sources | Prepare a report or draft and save it to the relevant task’s Work products. |
| Spreadsheet follow-up | Read selected rows, map recipients and prepare messages | Use the configured mail action, then update the row only after a successful outcome. |
Start, pause and stop
| Control | What it means |
|---|---|
| Start | Enable eligible runs in the selected work area under its saved map. |
| Pause | Retain the configuration while preventing further scheduled work. |
| Stop | Interrupt execution and prevent it continuing. Work already completed in an external app is not undone. |
| Manual run | Run outside the schedule when explicitly requested. Inspect the result before repeating external writes. |
Read execution history, not just the chat reply
The schedule shows when the agent is next eligible to run. A queued item is waiting; it is not a completed action. A check with no matching work is different from a task completed by the agent.
Inspect the run, individual step result and any saved work product. A drafted message is not a sent email, and a proposed map change is not applied until the interface confirms it has been saved.
Permissions and live data
The account, project assignment, connected-app authorization and configured approval policy bound execution. A connection can exist while a specific resource or operation remains unavailable.
Saving a reusable route does not mean its future tasks already exist. During execution, the route must handle empty inputs and missing data explicitly. A stopped agent cannot reverse an email already sent.
When a run needs attention
- 1Open the failed step
Read its specific reason: missing input, expired authorization, unavailable resource, permissions or a provider error.
- 2Correct that dependency
Select the right resource, reconnect the app or adjust the input/condition. Do not replace missing facts with invented data.
- 3Check what already succeeded
Completed external actions may remain completed. Resume or retry from the controls provided for the run.
Company knowledge follows the agent’s permitted scope
An agent can use company knowledge only within the current company, work and enabled source-action boundaries. A run may use live workflow inputs without retrieving saved memory. The company memory map reports recorded use; a successful run does not prove that stored memory was used.
To stop an agent’s workflow, use its workflow controls. Pausing company memory collection stops collection and memory retrieval; it is not a universal stop for all operational actions.