Using Deep Reasoning
Use Deep Reasoning Mode when precision and transparency in multi-layered analysis are more important than speed.
Where to Find It
The Deep Reasoning Mode toggle is located at the top-right corner of the chat interface. You can enable or disable it on a per-question basis during a session, allowing you to choose when deeper analysis is necessary.
What It Does
When enabled, Deep Reasoning Mode activates multi-intent decomposition for complex problem solving. The system:
Breaks down the main question into a graph of interdependent sub-intents.
Generates autonomous semantic graph for each sub-intent, allowing for rich contextual exploration.
Uses these semantic graph to build a structured understanding of the problem, improving interpretability and precision.
This is especially useful for queries involving compound logic, ambiguous phrasing, or multi-step analytical chains.
Impact & Things to Consider
Increased Analytical Depth: Produces more thoughtful, nuanced answers for complex or layered questions.
Higher Latency: This mode introduces additional compute overhead, so responses may take longer to generate.
Flexible Control: You can toggle it on or off for each question, depending on the level of reasoning required.
Inspector Behavior in Sub-Intent Nodes
The 'Inspector' - which reveals detailed metadata for entities in the semantic graph on double clicking any node - is currently available only for entities in the semantic graph of the main intent i.e. the root node, instead of the sub-intents.
For sub-intent nodes, the Inspector is disabled by design to keep these graph streamlined and easier to navigate.
Clicking on entities in sub-intents will not open the Inspector.
If deeper details about an entity are needed, navigate back to the root node’s semantic graph where inspection is fully supported.
Last updated