Guide

How to approve AI agent actions by text message

To approve an AI agent action by text message, you connect the agent to TextAlerts.ai, which texts you only when the agent hits a point that needs a human decision — then you reply approve (or reject, assign, snooze, or escalate) and the action is routed back to the system that asked. The SMS becomes the human-in-the-loop approval channel for your agents, so an autonomous agent can keep working but a human stays in control of the decisions that matter.

The approval loop, step by step

  • Connect your agentPoint your autonomous agent (Kane, Midas, HAMR, MiniCFO, or your own) at TextAlerts.ai. Every signal it produces is ingested into one command layer.
  • Let it score the signalTextAlerts.ai scores each event for urgency and whether a human decision is actually required, rejecting the noise — typically more than 90% of inbound notifications.
  • Get texted only when you're in the loopWhen an agent action genuinely needs sign-off, you get one concise SMS with context and a next-best action — not a notification firehose.
  • Reply to approve (or reject, assign, snooze, escalate)Reply with a number or command. The mapped action is routed back to the originating system, so approving runs the action and rejecting stops it.
  • Everything lands in the audit ledgerEvery alert and every reply is written to a full audit ledger, so each agent action has a record of who approved it and when.

The reply commands

Each alert offers a subset of these, depending on the rule. You reply with a number or the command itself:

  • approveproceed with the recommended agent action.
  • rejectdecline; nothing happens.
  • assignhand the decision off to a teammate.
  • snoozedefer and re-alert later.
  • escalatepush to a higher-priority channel or person.
  • customtrigger a workflow you've defined.

How this differs from alerting and SMS tools

TextAlerts.ai is an approval / human-in-the-loop layer, not an incident escalation tool and not raw SMS plumbing. Incident tools like PagerDuty and ilert are built to wake an on-call human for a failing service; SMS APIs like Twilio and SignalWire are the transport you would build on top of. TextAlerts.ai sits above both: it decides when an agent action needs a human, and maps your reply back to the action.

TextAlerts.aiApproval / HITL layerPagerDuty / ilertIncident escalationTwilio / SignalWireSMS plumbing (API)
Primary jobDecide when a human must approve an agent action, and route the reply backGet the on-call human to acknowledge and resolve an incidentSend and receive SMS programmatically
Reply maps to an actionYes — approve / reject / assign / snooze / escalate routes back to the originating systemReply acknowledges or resolves the incidentNo — you build the reply-to-action logic yourself
Built for autonomous AI agentsYes — purpose-built as the human-in-the-loop channel for agentsNo — built around services, monitors, and on-call rotationsNo — it is a messaging transport, not an agent layer
Signal scoring before it texts youYes — scores each event for urgency and human-decision-required, rejecting the noiseRouting and alerting rules, oriented to incidentsNo — sends whatever you tell it to send
Audit ledger of decisionsYes — every alert and reply is logged for agent accountabilityIncident timelines and post-incident reviewMessage logs only

Comparison reflects each tool's primary job, not full feature parity. TextAlerts.ai is in private beta.

TextAlerts.ai is in private beta. Join the waitlist to wire it into your agents.