The agency pattern is one Workspace per client. The same organizing idea can represent a project, department or other isolated operating environment. This describes what the Workspace represents, not an additional enterprise administration or migration feature.
How is a Workspace different from a chat, folder, model or computer?
| Object | Its role | What it does not establish alone |
|---|---|---|
| Chat | A conversation about work | Approved access to the correct client’s systems |
| Folder | Stores files; a canonical folder can anchor a Workspace | Approved identity, accounts and execution boundaries by itself |
| Model / runtime | Reasoning and execution tools | A durable client identity across tools |
| Computer | The place execution happens | Which client account or resource the task may use |
| Client Workspace | The client’s approved operating context and work | Universal migration or identical support in every runtime |
Can AI agents switch models to improve their economics?
A supported runtime choice can become more valuable when a lower-cost option meets the workload’s needs. Keeping client identity, Context, Skills, approved access and durable work in the Workspace makes that setup the starting point for evaluating another engine. It does not eliminate configuration or validation.
In Stable 1.45.0, Routines retain instructions, schedule and history between ready Claude Code and Codex runtimes. This is specific continuity, not universal session migration. Smaller, open and local models may offer further choices as their suitability and support are verified. Outloop does not route automatically by price or guarantee reduced inference costs. Compare the full cost per workload.
What can be reused—and what stays private?
Reuse reviewed client-neutral methods, Skills, workflow/Routine logic and supported integration patterns. Later implementations do not start from zero. Configure each Workspace’s own identity, Context, approved accounts/resources, access and browser mode; validate it separately.
Private client Context, client-specific Learning, files and work history do not automatically move between Workspaces. See the first-implementation versus next-Workspace workflow.
Why give each client a separate Workspace?
An agency may use the same AI engines for several customers. The client’s brief, Google Ads account, Meta account, CRM and Drive resources still differ. A successful login is not enough: the task must reach the intended client resource.
Client AWorkspace
- Context
- Client A’s brief & rules
- Accounts
- Google Ads A · Meta A
- Resources
- CRM A · Drive A
- Browser
- Approved identity A
- Access
- Approved for Client A
Work stays with this client.
Client BWorkspace
- Context
- Client B’s brief & rules
- Accounts
- Google Ads B · Meta B
- Resources
- CRM B · Drive B
- Browser
- Approved identity B
- Access
- Approved for Client B
Work stays with this client.
Follow the wrong-client prevention guide, then see why one shared login still needs resource boundaries.
A Workspace keeps more than a conversation.
Context gives work its client brief. Skills provide reusable methods. Reviewed Learning preserves approved lessons. Routines turn repeat work into a task you can run again—with its history intact.
Explore what belongs to a Workspace →- Context & Skills
- The client’s approved knowledge and methods for doing the work. Adding knowledge does not grant access.
- Reviewed Learning
- A person reviews proposed lessons before they change approved Context. Private client knowledge is not automatically transferred.
- Routines & history
- Run on demand or on a schedule. Review revisions, run results, activity and changed files. Switch a Routine between ready Claude Code and Codex runtimes while keeping its instructions, schedule and history.
Routines are available in Stable 1.45.0 and start only after you enable them on the Mac. Runtime readiness and supported capabilities apply.
What belongs in the current Outloop Workspace?
- Client identity and root: one canonical client folder, used across supported runtime integrations.
- Approved Context, Skills and reviewed Learning: client knowledge, working methods and reviewed changes.
- Accounts and resources: the systems and targets approved for that client.
- Browser access: approved identities and supported Browser Login recovery; Local Chrome uses the Workspace’s configured profile, while Managed Browser uses a shared profile with approved identity and access.
- Routines and records: instructions, schedules, revisions, run history, activity and output files within the verified runtime scope.
- Working environment: files and supported execution tools on the company-controlled Mac.
What is proven when a Routine changes runtime?
Stable 1.45.0 supports Routines with Claude Code and Codex after readiness checks. Installed QA ran the same Routine with both engines and retained its identity and history. The instructions and schedule belong to the Routine, rather than to either engine.
Read the Stable 1.45.0 release notes. Support for an engine in other workflows does not imply identical Routine support. Hermes and other integrations must be evaluated at their own supported scope.
What does “Models change. The Workspace stays” not promise?
It does not promise automatic transfer of conversations, live browser sessions or running tasks across arbitrary providers and computers. Broad portability is a product direction; it is not a universal migration feature available today.
Current release notes determine feature availability; a roadmap or another runtime’s features do not expand the supported scope.
Should the Workspace run locally or in the cloud?
Separate model location, execution location and client control. Outloop currently runs on a company-controlled Mac; a cloud model can still process task content under its provider’s terms. Read the local versus cloud decision guide and current Product overview.