Connection availability
Understand the difference between an AI interface, a configured model provider, and a connection enabled for your organisation.
Your Twin can be used in different places, but availability is configured per organisation and deployment.
Current documented routes
| Where you work | How the Twin is used | Availability boundary |
|---|---|---|
| Twin workspace | Full Sessions, Channels, files, views, and shared context | Available through your organisation's workspace |
| ChatGPT | Hosted app or connector using protected sign-in | A workspace owner enables the app; each user signs in |
| Claude | Hosted custom connector using protected sign-in | Personal setup or Team/Enterprise owner approval; custom connectors remain beta |
| Codex | User-scoped remote Twin connection for local development | Enabled through the organisation's connection pack |
| Claude Code | User-scoped remote Twin connection from a coding terminal | Enabled through the organisation's connection pack |
| Your own software | Governed queries and explicitly scoped channel operations through REST or the TypeScript SDK | Supplied for managed implementations when the route is enabled |
| Twin command line | Channel discovery, queries, and user-approved channel operations from a terminal | Public GitHub Release; Twin access is granted separately through browser sign-in |
All of these are adapters into the same governed Twin. None creates another copy of the organisation's business context or lets the client choose arbitrary data, SQL, models, or internal tools.
An interface is different from a model provider
ChatGPT, Claude, Codex, and Claude Code are places where a person or agent may use the Twin. The model that performs a configured task is a separate deployment decision.
Gemini can be selected as a configured model provider in a deployment. That does not mean the Gemini consumer product has a general, self-service Twin connector. No person-managed Gemini connection route is currently documented here; ask the implementation team whether a particular Gemini surface is enabled before describing it as available.
This separation lets the Twin retain the same reviewed business context while an approved model route changes after the required evaluation and controlled rollout.
Enabled is different from available in the product
A named connector or provider may be implemented without being configured for your organisation.
Treat a route as ready only when:
- your connection pack names it;
- the required owner or administrator has enabled it;
- your protected sign-in resolves to the intended organisation;
- the intended Twin context and channels are active and accessible for that route;
- a real governed question completes with the expected answer shape and any permitted channel links; and
- the provider's own data and retention controls have been approved by your organisation.
Ask for the route, not just the provider name
“We use Claude” or “we use Gemini” does not establish how that product reaches the Twin, which model handles a task, or what data controls apply. Your connection pack records the configured path.
Continue with Choose where to work when you know which route your organisation has enabled.