An always-on agent is an AI agent that keeps running in the background without a user initiating a conversation. It wakes when an external event arrives, such as a new email, a payment notification, or a webhook call, or when a scheduled time is reached; it then performs a short unit of work, saves its state, and returns to an idle state. Despite the name, the model is active only for brief periods and spends most of the time dormant.
Unlike chat assistants that respond only to user input, or stateless scheduled jobs that run at fixed times, an always-on agent acts proactively on events and timers and keeps persistent memory between runs. Building one reliably typically involves an event queue, timers for schedules and follow-ups, durable memory, permission controls, and rate control over notifications sent to the user. Because the agent acts autonomously while the user is away, credential isolation and human approval for high-risk actions are considered essential.