Skip to main content

Supervisor

Drives several specialized AI Agent nodes (in supervised mode) connected to its "agents" input, runs them in parallel on the document, then merges/deduplicates/arbitrates their annotations to produce a consistent final result — never inventing new ones. See Supervised Pipeline for details on the reconciliation mechanism.

Ports​

  • "Content" input (left) — the main document.
  • "Index" input (top, optional, document index type) — accepts only a single source.
  • "Agents" input (bottom) — accepts several AI Agent nodes in supervised mode simultaneously.
  • Output (right, annotation type).

Configuration​

  • Arbitration AI model — used only to settle conflicts between annotations; optional (defaults to the first active model). It is recommended to use one at least as capable as the models of the agents it supervises.
  • Severity:
    • Conservative (default) — only drops an annotation in case of a demonstrable false positive or an exact duplicate; keeps it when in doubt.
    • Strict — also arbitrates partial overlaps (keeps the more precise one, or merges them); still keeps it when genuinely in doubt.
  • Maximum number of arbitration calls — default 20, 0 = unlimited. Beyond that, remaining annotation groups are kept unchanged.
  • Additional instructions — optional free text.
  • Use support documents — unchecked by default; injects support documents' text into the supervisor's context.