A run is one execution of an installed agent. Dokki currently records two run origins: a conversation with the agent and a scheduled task.
Chat runs
Select an installed agent in Copilot or use the routing composer to start its dedicated conversation. The agent uses its current instructions, allowed workspace reach, enabled skills, pinned resources, memory setting, and configured MCP servers. A conversation can continue across turns; the activity timeline records the latest user prompt and a compact output preview for each run.
Scheduled runs
Schedules are configured on the agent's Overview tab. You can create and edit a schedule's name, task, cadence, timezone, target workspace, and enabled state. The panel also shows the next run and the latest run status. They are suitable for predictable recurring work such as weekly reviews, research briefs, or documentation checks.
You need Editor or Admin access to the target workspace, and the agent must have write access there. Changing the cadence or timezone recalculates the next occurrence. Disabling a schedule clears its next run without deleting its configuration.
Every scheduled occurrence starts without reading the agent's earlier conversation history. Its task still uses the agent's current instructions and configured capabilities, but the schedule prompt itself must contain the scope, inputs, expected output, and destination needed for that occurrence.
When a due schedule is claimed, Dokki creates a queued agent run and calculates the next occurrence. If its previous scheduled run is still queued or running, Dokki skips the new occurrence instead of running the same schedule concurrently. Scheduled work does not run in archived workspaces, and a paused or deleted agent cannot continue as an active worker.
Review the activity timeline
The Overview timeline is the record of work. It shows whether each run was started by chat or a schedule, along with status, model, step progress, timestamps, the triggering prompt preview, output preview, and an error message when one is available. Open the relevant conversation for the full exchange; use the timeline to diagnose repeated failures or an unexpected schedule result.
You can create, edit, enable, disable, or delete schedules from the Overview tab. Disabling preserves the schedule configuration but prevents future runs; deleting removes that schedule. Pausing an agent and disabling a schedule are different controls, so use both deliberately when retiring an automation.
Operational checklist
Verify the agent's workspace allowlist before creating a schedule.
Keep recurring prompts specific about scope, expected output, and where results belong.
Watch the first run and inspect any error before treating a schedule as dependable.
Revisit the schedule after permission, model, or MCP connection changes.