Projects and agents are visible objects
Agent Center shows which agents belong to which projects. A project object is the working context; an agent object describes one bounded responsibility. Connecting them creates an explicit project-agent relationship rather than a global assistant with unrestricted access.
Active connections animate; paused agents remain visible without implying that scheduled work is running. Object positions and the viewport are saved without continuously writing every drag movement to the server.
Select an existing project before adding agents
- 1Open Agent Center
Use Extensions from an authorized cabinet. Agent Center does not create projects, so there is no New project action inside the field.
- 2Choose a project object
Select a real project available to the current account. Its name and saved state become the context for linked agents.
- 3Open Library
Search the smaller Agent objects list and choose a template whose responsibility matches the project need.
- 4Add the agent
Connect the agent to the selected project. The relationship begins with an explicit runtime state and its own settings.
- 5Configure before relying on automation
Open Settings, review focus areas and behavior, save the operating rules and confirm the report time or trigger model.
Choose an agent with one clear responsibility
| Agent | Primary responsibility | Useful when |
|---|---|---|
| Project Coordinator Agent | Prepare a simple internal or client project path and request approval for the next real action | You need help choosing a saved team, creating a team first or opening Architect/Canvas |
| Sprint Agent | Review sprint scope, dates, acceptance criteria and workload risks | A current task plan needs a delivery-readiness review |
| Project Advisor | Review project health across the brief, roadmap, team, tasks, capacity and commercial boundaries | The project owner wants evidence-based assessment, practical recommendations and verified actions under approval |
| Digital PM Twin | Run a controlled project-management cycle, collect missing status and surface decisions | An active partner-managed project needs recurring attention |
| Invoice / Billing Agent | Reconcile accepted work, open invoices and payment milestones | Commercial project evidence needs focused review |
| CRM Follow-up Agent | Identify and prepare the next client relationship action | Team Studio needs a focused follow-up draft |
| Expense Agent | Review project expenses, approval status and vendor context | A client or partner needs to resolve project spending |
| Team Builder Agent | Ask structured questions and prepare a reusable team draft | A partner wants to create a team from a guided conversation |
| Team Capacity Agent | Review project allocation, role coverage and overload risk | A managed project needs workload and ownership review |
Use start, pause and stop deliberately
| Control | Meaning | Scheduled work |
|---|---|---|
| Start | The project-agent relationship is active | Runs according to saved rules |
| Pause | The relationship remains configured but is inactive | Must not run |
| Stop | The active relationship is ended | Does not run |
Do not confuse runtime state with settings
| Control | Stored meaning | When it takes effect |
|---|---|---|
| Start | The project-agent relationship is active | Immediately after the persisted state change |
| Pause | Configuration is retained but scheduled execution is blocked | Immediately; unsaved settings do not override Pause |
| Stop | The active relationship is ended | Immediately for future agent work |
| Save settings | Focus, behavior, report time and special instructions are stored | The next eligible run reads the saved version |
| Open Dock | The selected agent conversation becomes available without closing Agent Center | Immediately; closing Agent Center does not discard the Dock session |
Read connection and activity status
- An animated connection means the relationship is active, not that the model is continuously generating output.
- A still connection with Paused means scheduled work must not run.
- Hover or focus an agent or connection to inspect the last update, recent activity and conditions that need attention.
- A project can use several agents only when their responsibilities are distinct; remove or pause redundant agents instead of creating duplicate notifications.
- The right-side inspector belongs to the selected object. On mobile, use the circular panel control to open and close it without covering the field permanently.
Agents remain inside project and account boundaries
Move an agent result into delivery deliberately
- 1Open the right project agent
The Dock inherits the selected project-agent relationship; verify the project before asking for a deliverable.
- 2Name the target task
A saved result must use a real task in the same authorized project rather than a free-floating chat label.
- 3Ask for the working output
Request the report, document, spreadsheet or other supported artifact with a clear completion standard.
- 4Confirm the save action
The authorized agent action stores the result in that task's Work products. A normal chat reply alone is not attached evidence.
- 5Review in Workspace
Open Task Details, preview or download the item, then approve it or send a concrete revision note.
Before you expect an agent result
- 1Confirm the project
Make sure the agent is linked to the intended project object, not another project with a similar name.
- 2Confirm state
The object and inspector must both show Active; Paused and Stopped agents do not run scheduled work.
- 3Confirm saved rules
Resolve Changes not saved before expecting new settings to influence the next cycle.
- 4Confirm evidence exists
A project with no tasks, dates, messages or commercial records cannot produce a detailed evidence-based report.
- 5Confirm report timing
Scheduled output follows the saved report time and cadence. Manual Dock questions can be asked separately.