Size subagent work and recover individual tasks
New agent runs can choose concurrency from ready work, sandbox capacity and the remaining model allowance.
Choose Auto or set a maximum per run. The Team tab explains the current selection and shows task dependencies,
assigned files, progress and saved results.
Agents opens on Runs with creation actions ready. The model menu lists names, supports custom IDs and stays scrollable on small screens. Invalid team sizes explain their bounds beside the field; Working now counts only tasks that have started. After cancellation, the run shows that it is stopping and hides controls until termination is confirmed. The original prompt stays visible during live updates. Execution attempts are counted separately from saved steps, so a startup retry does not claim that work was recovered.
Retry a known failed task, redirect it with new instructions, or cancel it while keeping completed peers and existing files. Changes apply between operations, and accepted requests display their application state. The same controls are available through CLI, Python, API and MCP. Parent verification remains required.
Recovery preserves accumulated usage and shared limits. Uncertain external effects remain fenced, and BYOK provider spend is not measured or capped by Nodus. These controls apply to new runs; existing runs keep their recorded behavior. Local UI previews are explicitly labeled and do not display a running billing clock.