Connect Anthropic

In Settings → Model Provider, choose Connect Anthropic. Anthropic opens in another tab. Complete sign-in there, then paste the returned sign-in code into the Anthropic card and press Connect. If the tab did not open, use Open Anthropic. Terminal is not required.

The card loads models and sends one tiny request using your account to verify that Claude can reply. This uses a small amount of provider usage. It has no tools, skills, MCP servers, external delivery, or workspace files. On success, your existing available model is retained or the default available model is selected. Choose Open Alshival to return to chat.

Administrators use the same flow for shared, Team, and background connections under Advanced connections. Connecting verifies that account; it does not change another workload's selected connection. Other members cannot submit codes or cancel an administrator's sign-in.

Reconnect and retry

If Anthropic revokes or expires access, chat and Settings show Reconnect Anthropic. Existing credential files alone do not clear that warning. Complete sign-in again; your conversations are retained. Retry message restores the failed text to the composer for review and sending. It never automatically replays a turn or its commands.

Usage limits, unavailable models, and temporary provider problems offer a connection-check retry instead of requesting another sign-in. Verification runs once after login; page refresh and status polling do not consume repeated test requests. A sign-in attempt expires after ten minutes. Cancel or start again to replace it. Reloading Settings resumes a still-active attempt owned by you.

The native CLI stores credentials only in the selected connection's isolated account home. Sign-in codes are never placed in chat, browser storage, audit records, or Terminal history. See the architecture decision.