Offline use
Offline use is possible, but it is a property of where the model runs rather than a setting you switch on.
What works offline
Local AI, once the runtime and the model are already on the machine. The request goes to loopback and never leaves it:
bccli -m provider/modelThe install, the model files, and the runtime have to be in place before you disconnect — nothing is fetched at request time.
What does not
BotConnector Cloud requires internet access. So does anything that explicitly reaches a network: web search, remote tools, cloud model IDs, and the Workspace screens you use to manage keys or billing.
That distinction is worth knowing before a flight or a site visit: a session pinned to a local model keeps working, while one pinned to a cloud model stops with a network error, and both look identical until the request is made.
Planning around it
If you need a session that survives a disconnection, choose the local provider up front rather than switching mid-task. Switching is easy while the network is up and impossible once it is not.
Next
- Use Local AI in BCCLI — pinning the local provider.
- Network boundary — which addresses each surface uses.