Docs

Scheduled runs

Ask an agent to do something later. How schedule_run works from chat and MCP, and how the result comes back to you.

A scheduled run invokes your agent later, with a prompt you set now. "Check the release issue tomorrow at 9" becomes a real run that fires at 9, works on the agent's own computer, and posts its reply back to you.

Two ways to schedule

Ask the agent in chat. plori agents can schedule themselves. Say "every detail is in NOTES.md, pick this up tomorrow morning" and the agent registers the deferred run on its own. This is the everyday path.

From your own tools. The MCP server exposes a schedule_run tool, so Claude, ChatGPT, or your own scripts can schedule work on a plori agent:

{
  "name": "schedule_run",
  "arguments": {
    "agent_id": "<your agent's id, from list_agents>",
    "prompt": "Summarize new issues since yesterday and triage them.",
    "fire_at": "2026-07-03T09:00:00Z"
  }
}

Give either fire_at (an RFC3339 timestamp) or delay_seconds (relative, from now), plus the prompt to send. A time in the past is rejected.

What happens when it fires

The schedule fires as an ordinary run: the agent wakes if it was asleep, receives your prompt, and works with its usual disk, tools, and memory. Credits are drawn the same as any run; if your balance is empty, the schedule is marked failed rather than silently dropped.

The reply is delivered where you will see it:

  • If you have a conversation with that agent open, the reply arrives inline there.
  • Otherwise it lands in a new thread named "scheduled", waiting in the agent's history.

You see only the agent's report, not a re-echo of the scheduled prompt.

Scope of v1

Scheduling is currently one-shot: each schedule fires once. For recurring work, the practical pattern today is to end a scheduled run by asking the agent to schedule the next one, or to drive the schedule from your own cron via MCP.