Unattended work on plori
Background agents that keep working.
A background agent is an AI agent that continues a task after you stop watching it. It runs in the cloud, not in your open chat or editor, and reports back when the work ends. Most tools that use the term mean a coding agent that works on a repository and returns a pull request. plori's background agent is a general-purpose agent with its own persistent environment. It continues accepted work after you close the tab: research, scripts, files, or builds. You can also use it for code.
Opens your agent straight away, no signup needed.
What is different about plori's background agent?
§01A background coding agent usually starts from a repository and a task. It clones the code into a cloud sandbox, makes the change, runs the tests, and gives you a branch or a pull request to review.
plori provides the agent. You do not deploy your own. Each agent has a persistent disk, a shell with tools and web access, and saved memory notes. The work can be code, but it can also be a research report, a data script, a scrape, or a file that the agent builds for you. The result is a reply in the conversation and the files on the agent's disk. The persistent environment page explains what stays on that disk.
Which kinds of unattended work does plori run?
§02Background tasks
When a command runs for more than ten seconds, the agent moves it to the background and ends its turn. One task can run for up to one hour, and one agent can run up to five tasks at once. When a task finishes, the agent posts one summary in the same thread. The full output stays on the account disk. See background tasks.
Scheduled runs
A scheduled run starts a new agent turn at a time that you set. Ask for it in chat, or call schedule_run through MCP. Each schedule fires once. See scheduled runs.
Standing workflows
For the same steps every time, the agent builds a workflow that fires on a schedule or a webhook. A workflow run is not an agent turn, so it does not wake the agent. Every run stops at a five-minute safety limit. See AI-built workflows.
Approval and input requests
When a run needs an approval or a secret, it pauses and waits for your answer. plori collects secrets in a masked input and removes them from the visible history.
What can stop unattended work?
§03Closing the tab or shutting your laptop does not stop accepted work. These things can stop it or pause it:
- a runtime failure or a restart
- a limit: one hour for a background task, five minutes for a workflow run
- an approval or input request that nobody answers, because the run waits
- an empty balance: a scheduled run is then marked failed, not dropped without a record
plori is not an always-on process host. An idle agent sleeps. Sleep keeps saved files and memory notes, but it does not keep running processes or RAM. The agent does not go to sleep for inactivity while a background task runs. A failure or a restart can still end a process, and plori does not resume that process from the same instruction. For long work, tell the agent to write checkpoints to disk. A later run can then continue from the last saved file.
How can I try a background task?
§04We have not published a recorded run of this task. Each step states behavior from the background tasks guide. Use a command that takes more than ten seconds, for example the test suite of a repository.
- Open an agent at plori.ai/agent, or connect Claude Code with one prompt: Set up https://plori.ai/SKILL.md. The Claude Code example records a verified run of that connection.
- Send this task, with your own repository in place of OWNER/REPO: Clone https://github.com/OWNER/REPO, install its dependencies, and run the test suite.
- Wait for the reply. If the suite runs for more than ten seconds, the agent moves it to the background and tells you that the job is running.
- Close the tab.
- Open the conversation again later. The agent posts one summary in the same thread, with the tail of the output. Ask the agent for the full log.
The connection check on the example page uses no credits. The test run uses credits for model use and compute. Check pricing before you start.
When should you use something else?
§05- Your coding tool already runs agents in the cloud or on a schedule, and the task ends in a pull request. Compare that tool first. The comparison guide covers the tradeoffs.
- The steps are the same every time. Use a workflow.
- The work is short and you watch each step. Keep it local. A short cycle of edit and test is faster in the editor that you already have open.
To send work from the agent you already use to plori, see MCP delegation. For repositories, builds, and tests, see the cloud coding agent page.
Questions, answered.
§06What is a background agent?
A background agent is an AI agent that continues a task after you stop watching it, and reports back when the work ends. Most products use the term for a coding agent that returns a pull request. plori's background agent is a general-purpose agent that runs accepted tasks in its own persistent environment.
Can a background agent work while my laptop is off?
Yes, after plori accepts the task. The work runs in plori's cloud, so closing the tab or the laptop does not cancel it. A runtime failure, a task limit, an unanswered approval request, or an empty balance can still stop or pause it.
How do results reach me?
A background task posts one summary in the conversation that started it, and the full output stays on the account disk. A scheduled run replies in an open conversation with the agent, or in a new thread named scheduled. Over MCP, get_run_result reads a run that finished later.
What happens when the agent needs approval or input?
The run pauses and waits for your answer. Approve or answer in the web app or in the CLI with plori attach. An MCP client can deny a request or answer a question, but only you can approve an action, in the web app. Connect the accounts and credentials that the task needs before you leave it unattended.
How do I set up recurring work?
Use a workflow when the steps are the same every time. It fires on a schedule or a webhook without an agent turn. A scheduled run fires once. For work that needs the agent's judgment each time, tell the agent to schedule the next run, or call schedule_run from your own scheduler.
Is a plori background agent always on?
No. An idle agent sleeps, and sleep keeps files and memory notes but not processes or RAM. A background task can run for up to one hour, and a restart or failure can end it sooner. Save checkpoints to disk for work that must continue.
What does a background agent cost on plori?
Agent work uses credits for model use, compute, and paid tools. Workflows have their own run charges, plus any model use in their steps. Sleep stops agent compute charges, but a paid subscription and extra disk still cost money. Check pricing for included allowances and current rates.
Is plori a background coding agent?
You can use it as one. The agent can clone a repository, run builds and test suites as background tasks, and use git. It is not built around a pull request, and it also runs work that is not code. If your coding tool already returns pull requests from a cloud agent, compare that first.