Chat is a fine way to start work and a terrible place to wait for it. A plori agent takes the job, keeps it running on its own computer after the reply, fires scheduled work while you sleep, and posts results back into the conversation.
Opens your agent straight away, no signup needed.
A slow build, scrape, or batch job moves to the background instead of hanging the turn, and the agent reports back in-thread when it finishes. How it works.
"Pick this up tomorrow at 9" becomes a real run that fires at 9, in chat or via the schedule_run MCP tool. How it works.
When unattended work hits a decision only you can make, the run parks, asks, and resumes with your answer. No silent guessing.
Background work only makes sense if the worker persists. An agent that evaporates with the browser tab cannot finish a two-hour job or hold the context for tomorrow's scheduled run. Every plori agent has a persistent computer with its own disk and memory, so unattended work lands on the same machine, with the same files, that your conversation was about.
And because idle agents scale to zero, an agent that works for twenty minutes a night costs you those twenty minutes, not a 24/7 server.
A background agent is an AI agent that keeps working when you are not watching: it runs long jobs past the end of a chat turn, fires scheduled work at a set time, and reports results back to you afterward. On plori this is the default shape of an agent, because each one lives on its own persistent cloud computer rather than inside a single chat session.
In the same conversation. A finished background job posts back into the thread it started from: if you have it open the reply streams in live, otherwise it is waiting the next time you look. A scheduled run lands in your open conversation with that agent, or in a new thread named scheduled.
Yes for scheduled runs: they fire on plori's infrastructure at the time you set, wake the agent, and save the result. That is the point of the agent having its own computer: none of it depends on your machine being on.
The run parks instead of guessing. plori has built-in human-in-the-loop: the agent raises an approval or input request, you answer it from the dashboard or via MCP (list_pending_inputs and answer_pending_input), and the run resumes exactly where it stopped.
Scheduling is currently one-shot: each scheduled run fires once. For recurring jobs, have the agent schedule its own next run at the end of each scheduled one, or drive schedule_run from your own cron over MCP.
Nothing while asleep, within your plan's disk. Agents scale to zero between jobs; you pay credits only for the time they actually work.