Wide Research

Evidence-heavy research with visible sources

Wide Research is the operating pattern for tasks that need many sources, browser evidence, PDFs, or cross-checking before a conclusion.

Source discipline

Research output should show sources and what was inspected. The agent should distinguish verified evidence from inference, and it should avoid presenting unvisited links as proof.

Composer intake

The chat home Wide Research mode exposes depth, source scope, and output settings. Submitting a prompt should ask North Star to draft an editable plan first, then gather approved public web, connected app, file, or PDF evidence according to that scope.

Child agents

North Star can delegate bounded research tasks to child agents. The parent thread should show queued, running, blocked, completed, and reported states.

Failure state

If browser, Composio, file, or PDF access is unavailable, the agent must say which source class is blocked and continue only with the approved sources that are actually reachable.