Sign inSign up free
Omnichannel inbox Visual job flows Characters Any model + fallbacks CRM actions & booking Knowledge bases Approvals & autonomy Test & Debug A/B testing Analytics Website chat widget Quote desk Custom tools MCP server Rate limits & queue AI appointment setter All features Pricing FAQSign up freeSign in
Approvals & autonomy

You decide how much your agents can do on their own

Start cautious and open up as trust grows. Every agent has an autonomy level: suggest everything, draft replies for approval, or run fully on its own. Pending actions wait on one Approvals page, follow-ups re-engage quiet contacts, and nothing goes out outside your hours.

Try it free How it works
The same three actions run immediately or wait for approval depending on the agent's autonomy: suggest, draft or autoSuggestDraftAutoAGENT WANTS TO…add_taghot-leadupdate_deal→ Qualifiedsend_reply“Thursday 3pm works…”autonomy gateRuns nowWaiting in Approvalsadd_tag · pendingupdate_deal · pendingsend_reply · pendingadd_tag · doneupdate_deal · donesend_reply · pendingadd_tag · doneupdate_deal · donesend_reply · doneSuggest: every action waits for a humanDraft: lookups, tags and deals run; sends waitAuto: fully autonomous, all on the record
3autonomy levels
1Approvals page for every inbox
9am–8pmdefault follow-up window
1reply per run, maximum
Deep dive

A closer look at Approvals & autonomy

01 · Autonomy

Suggest, draft or auto, per agent

In suggest mode every action the agent wants to take goes to Approvals. In draft mode tags, notes and lookups run, but anything that sends waits for a human. In auto mode the agent is fully autonomous. Switch any time without touching the flow.

  • suggest: every action needs approval
  • draft: sends need approval, everything else runs
  • auto: fully autonomous
  • Changeable per agent, instantly
A matrix of which actions run automatically or wait for approval under suggest, draft and autosuggestdraftautoSend replySend emailBook appointmentTag contactUpdate dealSearch knowledgeasksasksasksasksasksrunsasksasksrunsrunsrunsrunsrunsrunsrunsrunsrunsrunsSwitch levels any time, no republishGuard: at most one reply per run
02 · Approvals

Review, edit and approve in seconds

The Approvals page lists every pending action with its conversation. Read the draft, tweak the wording, approve or reject. Approved sends go out through the same rate-limited pipeline, and wait politely if a provider is throttling.

  • One queue across all inboxes
  • Edit before approving
  • Approve or reject with context
  • Approved actions respect rate limits
On the Approvals page a drafted reply is edited, approved and sent, and the next pending action moves upApprovals · 3 pendingMaya · SMSsend_reply · 2 min agoHi Maya! Thursday at 3pm works great, I'vepencilled you in. See youat our Leeds studio!then!EditRejectApproveDana · Gmailsend_email · Quote for 12 seatsLeo · Instagrambook_appointment · Sat 10:00Sent to Maya via SMS
03 · Follow-ups

Follow-up sequences that know when to stop

Named nudge sequences re-engage contacts who go quiet, each step written fresh by the agent from its prompt and sent inside business hours. The moment the contact replies, the sequence stops and the conversation picks up where it left off.

  • Named sequences with timed steps
  • Each nudge written fresh from its prompt
  • Stops as soon as the contact replies
  • Scheduled follow-ups shown right in the conversation
A follow-up sequence sends a nudge on day 1 and day 3, then stops as soon as the contact repliesSEQUENCE · “Quote not answered”Day 0Day 1Day 3Day 7Here's your quote:$480 for Friday.Prompt: check in,mention Friday slotJust checking in,Friday's still free!sent 10:02Prompt: offer adifferent dayWould Monday suityou better?sent 09:15Prompt: last call,keep it lightcancelledSorry, been busy! Monday would be perfectContact replied: sequence stopped, conversation resumesNudges are written fresh from their prompt and sent inside business hours
Under the hood

The details

Guard: one reply per run

An agent can never send more than one reply in a single run.

Guard: last message outbound

If the latest message is already yours, the run is skipped.

Prompt-injection guard

Inbound content is wrapped and treated as untrusted data, never instructions.

Business hours

Wait for opening, send anyway, or send a holding message now.

Hand-off

Escalate to a human teammate and stop the agent on that conversation.

Notes

Leave notes on threads for your team and for the agent's context.

Use cases

Who it's for

Rolling out safely

Start in suggest, graduate to draft, then auto once the approvals are all green.

Regulated industries

Every outbound message is reviewed while lookups and tags run automatically.

Revival campaigns

Follow-up sequences re-open cold leads during business hours only.

FAQ

Approvals & autonomy: common questions

Can I change autonomy without republishing?

Yes. Autonomy is an agent setting and takes effect on the next run.

What if an approved send hits a rate limit?

It stays pending and goes out as soon as the provider allows, without losing an attempt.

When do follow-ups send?

Inside the inbox's business hours, or 9am–8pm when none are set.

Can the agent hand a conversation to a person?

Yes. Hand-off stops the agent on that conversation and notifies your team.

Keep exploring

Works hand in hand with

Channels

Omnichannel inbox

Gmail, Outlook, SMS, WhatsApp, Instagram, cold-email tools and web chat, answered by one agent.

Explore Operate

Test & Debug

Sandbox conversations, a full debugger for every reply, and an AI diagnose-and-fix loop.

Explore Build

CRM actions & booking

Book, reschedule and cancel appointments; update fields, deals, tags, notes, tasks and workflows.

Explore
Get started

Put it to work on your own inbox

Connect an inbox, describe the job, test it in a sandbox and go live with exactly the autonomy you're comfortable with.

Sign up free See pricing