A conversation without borrowing your account.

We are extending the phonebook so you can find people, agents and connected services in one place. You can ask a connected service such as Linear for records through its reviewed read tools. SAINT answers from those records and names their source.

Our pilot implementation also gives an agent its own revocable message mailbox. You review the exact message before it leaves SAINT. The agent must check its mailbox through its own host, and its reply remains attributed conversation content. We have not yet verified consumer-account connections to Grok Bot or Muse. You keep their action approvals enabled in their own apps.

We are also connecting the private contact book to a view of your recorded commitments. You can see what you promised, what others owe you and what SAINT has taken on. These private links let their holder navigate between contacts and work until the original link expires. You can prepare an editable email follow-up or choose calendar time from a commitment. You review the proposal before we send or book anything. We show the action status beside the commitment, which stays open until you resolve it. Calendar proposals do not check availability or invite anyone. This implementation is entering private verification; we have not yet verified it with live provider actions.

We have added an edit option for an unapproved follow-up or calendar proposal. Saving a revision withdraws the earlier proposal. You must approve the replacement before it can run. If you already approved the original, we ask you to check its result first. We still need live provider evidence for the complete coordination journey.

Mailbox drafts expire after ten minutes. Confirmed exchanges expire after seven days. You can remove a connection and its SAINT-held exchanges, but the provider controls its own copies. If you ask SAINT to discuss a reply, we retain that discussion under our normal conversation rules.

A person can say “talk to Claude”, “talk to ChatGPT” or “start Direct” by text. SAINT opens Direct and says who is answering. It does not claim access to an existing Claude or ChatGPT chat, project, subscription or provider memory.

Claude may use SAINT's approved Anthropic route. ChatGPT remains unavailable until an OpenAI route has separately passed the privacy and operational gate. A named provider cannot silently become another provider during fallback.

Direct loads no SAINT memory, connected sources, standing decisions or action tools. Its active window is limited to twelve messages and expires after two hours. Saying “back to SAINT” removes it immediately. The settings page can turn Direct off or choose a preferred provider, but it cannot weaken the privacy and memory boundary.

This is conversation, not delegated completion. Provider words are not evidence, consent, memory or a receipt, and time spent talking does not count as work SAINT completed.

“Hand this to SAINT” does not switch modes. It first shows the exact bounded transcript that would cross the boundary. Only a second “confirm handoff” removes the temporary provider session and gives that text to the ordinary secretary service.

The feature is text-only and limited to explicitly enabled private-pilot seats. Voice parity and short-lived continuation on signed-in consumer sites remain in development.