Docs

Background tasks

Long-running work keeps going after the reply. How plori backgrounds slow tool calls and posts the result back into the conversation.

Some work is too slow for a chat turn: a long build, a big download, a batch job. On plori, a slow command does not force the conversation to hang. The agent moves it to the background, answers you, keeps the work running on its computer, and reports back in the same conversation when it finishes.

What it looks like

  1. You ask for something that involves a long-running command.
  2. The command starts. If it finishes quickly (within roughly ten seconds) everything stays inline and you just get the answer.
  3. If it keeps running, it is moved to the background. The agent ends its turn and tells you the job is running.
  4. You can keep talking to the agent, or close the conversation entirely.
  5. When the job finishes, its result is posted back into the same conversation: the agent picks up the output, summarizes it, and replies. If you are looking at the thread, the reply streams in live; otherwise it is waiting the next time you open it.

There is nothing to configure and no polling to write. Delivery is exactly-once: a result is never dropped and never posted twice.

What runs in the background

Anything the agent can run in its shell: builds, test suites, scrapes, data jobs, installs. Background tasks have generous but real limits: a task can run for up to about an hour, and each agent can only keep a handful running at once. Work that should start later rather than run longer belongs to scheduled runs instead.

Where the output goes

The full output stays with the agent on its disk, so nothing is lost if the output is large. The completion message carries the tail of the output, and you can ask the agent to dig into the rest, since it is sitting in its files.