Skip to content
Book a demo

AI governance

An AI CRM that won't email without your approval? Where SAGARIS draws the line

In SAGARIS, the AI agent cannot send an email: it can only propose one for a person to confirm. A reply the AI drafts waits in your inbox until someone sends it; Inbox Auto-Mode, which would send low-risk replies on their own, is built but not offered to customers yet. What does send without a click per message is a running sequence, including the steps SAGARIS personalized for each contact, and a few confirmations and reminders listed below.

Checked against the SAGARIS product code on 25 September 2026; the opt-out rule as changed on 10 October 2026.

Nothing reaches a contact until a person presses Send or queues it.

How an email reaches a contact in SAGARIS, and what a person does first

  1. The AI agent: cannot send

    Waits for a person

    Its email, enroll and book tools are proposal-only. No autonomy level changes this.

  2. Replies drafted in the inbox: wait for a person

    Waits for a person

    The draft goes nowhere until a rep sends or schedules it.

  3. Inbox Auto-Mode: built, not offered yet

    Not offered yet

    When offered: admin opt-in, low-risk email or text replies only, classifier at least 80 percent.

    Built but not offered to customers yet, so every drafted reply waits for a person.

  4. Sequences: send on schedule once running

    Sends on schedule

    A contact enrolled in a running sequence gets its email and SMS steps on schedule without a click per message, including the steps SAGARIS personalized. A LinkedIn step is a task for your rep.

  5. Out-of-office follow-ups: part of a running sequence

    Sends on schedule

    A follow-up on the return date, a second a week later if there is no reply.

  6. Meeting confirmations and reminders: templates

    Sends on schedule

    Templates, not AI-written: reminders 48 hours, 24 hours and 1 hour before a meeting.

  1. The AI agent: cannot send

    The agent's tools that would email a contact, enroll one in a sequence or book a meeting are proposal-only. There is no send path behind them. The outbound switch and the suppression list are checked, then the tool returns a proposal marked as not sent, and a person has to act on it where sending actually happens. No autonomy level changes this.

  2. Replies drafted in the inbox: wait for a person

    When a contact replies, SAGARIS can draft the answer. The draft goes nowhere until a rep sends or schedules it, and the record keeps which person did.

  3. Inbox Auto-Mode: built, not offered yet

    Auto-Mode would let low-risk replies go out by themselves. It is built but not offered to customers yet, so today every drafted reply waits for a person. When it is offered, it will be off by default, only a workspace admin will be able to switch it on, and switching it on will ask for confirmation.

    Even then, a reply would only send by email or text, and only when the classifier is at least 80 percent confident and at or above the workspace's own threshold, the draft passes the quality check and makes no claim that cannot be verified, and the incoming message shows no sign of an attempt to hijack the model. Auto-Mode would never send a LinkedIn reply.

  4. Sequences: send on schedule once running

    A contact enrolled in a running sequence gets its email and SMS steps on schedule without a click per message, including the steps SAGARIS personalized for that contact. Call, manual and task steps send nothing themselves: they give a rep the work, and the sequence moves on if it is not done. A LinkedIn step is a task for your rep.

    Contacts reach a sequence in several ways, among them a person enrolling them, your own integration through the API, an enrollment rule someone created for buying signals, an inbound lead routed to a fast-track sequence, a won deal starting customer onboarding, and the Re-Engagement sequence SAGARIS manages for a contact who went quiet at the end of a sequence and later showed interest.

  5. Out-of-office follow-ups: part of a running sequence

    If a contact in a running sequence answers with an out-of-office that names a return date, the AI writes a follow-up and SAGARIS sends it on that date, sends a second a week after the first went out if there is still no reply, and then returns the contact to the sequence.

  6. Meeting confirmations and reminders: templates

    Someone who books through a SAGARIS booking link gets a confirmation, invitees can get reminders 48 hours, 24 hours and 1 hour before a meeting, and marking a meeting as a no-show queues a rescheduling email to the prospect. These are templates rather than AI-written text. Once a number is routed to the AI receptionist, a confirmation for a caller it recognizes goes out without a rep; for a caller it cannot identify, a rep approves the request before the caller is emailed.

The checks in front of an automatic send

A send that needs no click still has to get past these.

  1. Suppression, per channel

    An opt-out is keyed on the email address and the phone number rather than on the contact record, so it survives an edit, a delete or a re-import. A reply or a STOP stops the channel it arrived on; a Do Not Contact, a spam complaint or a request to leave everything stops every channel. If the suppression list cannot be read, the contact counts as suppressed.

  2. An emergency pause

    Holds sequence email, text, LinkedIn steps and scheduled calls for twenty-four hours, then lifts on its own. An unreadable pause state counts as paused.

  3. A master outbound switch, and where it stands

    Closed by default in code. Open on the SAGARIS production service, because the product sends real email, so it is not what stands between your contacts and a send. The paths and checks on this page are.

  4. Daily limits per sender

    Email from a SAGARIS-managed mailbox claims that mailbox's daily capacity in one step; on error the email waits. Sequence texts rotate across approved numbers and skip one at its daily cap.

How SAGARIS protects customer data

What this page does not claim

  • Not every email is approved one by one

    A running sequence does not ask again before each email; launching it was the approval. The AI agent proposes; drafts wait for a person.

  • Arming Auto-Mode is not in the audit trail today

    Auto-Mode is not offered yet. Its setting is stored, but a change to it is not written to the configuration audit trail, so the trail would not tell you who switched it on or when.

    What the audit trail records

Questions buyers ask next

Most asked

More questions? Book a demo
  • Yes: in SAGARIS you approve a sequence once when you launch it, and its email and SMS steps then run on schedule without a click per message, including the ones the AI personalized for each contact, while the AI's own proposals and the replies it drafts wait for a person. The AI agent's send, enroll and book tools are proposal-only, so it can suggest an email but a person confirms it, and a reply the AI drafts waits in your inbox until someone sends or schedules it.

    Inbox Auto-Mode, which would send low-risk replies on their own, is built but not offered to customers yet. When it is offered, it will be off by default, only a workspace admin will be able to switch it on, and even then it would send only low-risk email or text replies that the classifier is at least 80 percent confident about and that clear the workspace's own threshold and checks. If you want a person to write or check a particular message, use a manual or task step in place of an email step.

  • No. The agent's tools that would email a contact, enroll one in a sequence or book a meeting are proposal-only, with no send path behind them. It returns a proposal marked as not sent, and a person acts on it where sending actually happens. No autonomy level changes that.

  • No. A drafted reply waits in the inbox until a rep reads it and either sends or schedules it, so the AI writes and a person decides. Nothing leaves the workspace on the strength of a draft alone. Inbox Auto-Mode, which would send low-risk replies on their own, is built but not offered to customers yet.

    When it is offered, it will be off by default, only a workspace admin will be able to switch it on, and even then it would only send low-risk replies by email or text that clear an 80 percent confidence floor, the workspace's own threshold and the quality and claim checks. It would never send a LinkedIn reply.

  • No. Launching the sequence is the approval for its schedule, and its email and SMS steps go out on time without a click per message, including the steps SAGARIS personalized for each contact. Call and task steps send nothing themselves; they give a rep the work. A LinkedIn step is a task for your rep.

  • Yes. A manual or task step creates a task for a rep and sends nothing itself, and a LinkedIn step is a task for your rep. The sequence does not wait forever: if the task is still open a day after it is due, the step closes as not done and the sequence moves on.

  • The workspace emergency pause holds sequence email, SMS, LinkedIn steps and scheduled calls together, for twenty-four hours, and then lifts on its own rather than staying shut until somebody remembers it. If the pause state cannot be read at all, SAGARIS treats the workspace as paused rather than sending, so the failure direction is silence instead of an unintended send.

  • Partly. A sent draft records the rep who sent it, and an approved or rejected agent action records who decided and what happened. Auto-Mode is not offered yet, and switching it on would not be recorded in the audit trail today. The audit trail page lists what is recorded, who can read it, and what it does not cover.

Bring the emails your team sends and we will walk through which ones wait for a person.

Book a demo