Personalized Response Styles
Cortex doesn’t just answer — it learns how you like your answers. Over time, the system adapts to each user’s preferences by observing their behavior, past queries, and feedback loops. This includes:- Preferred formats (e.g. tables, bullet points, summaries)
- Preferred tone (e.g. concise, friendly, formal)
- Preferred depth (e.g. quick summary vs detailed explanation)
- Specific content types (e.g. Excel files, slide decks, contracts)
Smart Clarifications
When users ask vague or ambiguous questions, Cortex’s intelligent search agent recognizes this and asks for clarification instead of performing a poor-quality search. Rather than returning irrelevant or incomplete results from unclear queries, and just plain ChatGPT answers, the system:- Detects ambiguous language in user questions
- Identifies missing context that would improve search quality
- Generates targeted clarification questions to gather more specific information
- Waits for user input before proceeding with the search
Grounded with Citations
Every response can be grounded in source material using Cortex’s built-in citation system. When a user asks a question, the answer is generated using documents, web data, or memory — and the system:- Returns the exact chunks of source text used to build the answer
- Provides source IDs, titles, page numbers, and optionally bounding box coordinates for PDFs
- Allows the frontend to show inline citations, source previews, or even highlight regions in original files
Filter by Metadata
You can restrict the context used for answering questions by providingtenant_metadata
and document_metadata
parameters in upload API calls. This allows you to filter sources by source_title
or source_type
.
Creative Mode
Sometimes, users want answers that go beyond their knowledge base. Creative mode allows your app to let LLMs generate responses without solely relying on the retrieved context. It’s like interfacing directly with ChatGPT. Withcreative_mode
enabled, Cortex shifts from strictly grounded retrieval to imaginative generation, allowing it to:
- Suggest new ideas or concepts
- Generate content beyond the original sources
- Expand on partial context using LLM capabilities
- Brainstorming things outside of the context
- Writing summaries with flair
- Exploring hypothetical ideas
- Rewriting answers for tone and creativity
search_mode: ["creative"]
parameter in the /search/qna
API call.
Combine All Modes
Cortex’s strength is in combining these capabilities into one flexible response system:- Retrieval + Generation: Factual answers grounded in sources
- Memory + Personalization: Responses that evolve to match each user
- Creativity + Control: Creative responses when needed, factual when not