Thinker mode
Reasoning-first analysis. It uses the prompt assembly, supplied data, global context, and model reasoning without external web search.
AI Agent modes
Thinker mode and Researcher mode solve different forecast problems
Modes define how an AI Agent executes. Thinker mode reasons from the supplied context and data. Researcher mode can expand the evidence base when an asset requires additional external discovery.
Updated July 3, 2026
Overview
Mode is the execution layer
An AI Agent defines the analytical identity. A mode defines how that identity executes the task. This separation lets iPulse evaluate the same AI Agent across different capabilities without treating each execution style as a completely new advisor.
Researcher mode
Research-enabled analysis. It can look outward for additional current or narrow evidence when the prepared context is not enough.
What improved
Research is powerful, but it should not be the default for every task
iPulse treats mode selection as a methodological decision. Research-enabled execution can uncover valuable narrow evidence, but it can also introduce source variability, latency, cost, and inconsistent evidence depth across assets. Thinker mode stays useful because it keeps comparisons controlled when the supplied data and context are sufficient.
Controlled comparison
Thinker mode is useful for scaled asset sweeps where each agent should reason from the same prepared context, input format, and output schema.
Targeted evidence expansion
Researcher mode is better when a forecast depends on external facts that are too specific for the standard context bundle, such as lobbying, litigation, procurement, or new regulatory details.
Thinker mode
Best when the provided context is already sufficient
Thinker mode is useful when the forecast mainly requires structured reasoning from supplied data: price history, fundamentals, macro context, output schema, and the AI Agent framework. It is repeatable and controlled, which makes it useful for comparing many assets on the same footing.
Thinker mode does not mean shallow. It can use high reasoning depth, but it does not perform external web search as part of the generation path.
Researcher mode
Best when the missing evidence is outside the prompt
Researcher mode is useful when the answer may depend on narrow, current, or relationship-heavy facts not already in the prompt assembly. The prepared context may describe the macro regime, but it will not always include every current lobbying disclosure, management-government relationship, legal development, procurement contract, or regulatory detail for every company.
Researcher mode should therefore be interpreted as evidence expansion. It does not replace the structured prompt assembly; it adds an external discovery capability where the task calls for it.
Examples
Why certain AI Agents benefit from Researcher mode
Machiavelli: power structure
Researcher mode can look for lobbying spend, revolving-door relationships, procurement exposure, sanctions pressure, antitrust risk, and government links.
Sherlock Holmes: evidence audit
Researcher mode can look for filings, footnotes, related-party signals, allegations, or missing disclosures that may not appear in the prepared data.
Ray Dalio: macro regime
Researcher mode can cross-check current policy moves, central bank communication, and geopolitical updates against the supplied macro context.
Warren Buffett: ownership quality
Thinker mode is often enough for broad ownership reasoning, while Researcher mode may help when governance or management-specific facts need checking.
Product access
Mode details appear inside eligible prediction reports
The Config Details modal shows which mode was used for a prediction when the user has access to that prediction. Users must be signed in, and their subscription must unlock the relevant asset, batch, or advisor report.