# AI & LLMs Overview

# AI & LLMs Overview

Curiosity Workspace supports AI-assisted features that are grounded in your workspace data. In practice, this usually means:

• retrieve relevant information from search + graph
• apply LLMs to summarize, answer, classify, or orchestrate workflows
• optionally write results back to the workspace (notes, links, actions) via endpoints/tasks

# What to use LLMs for

High-value, common use cases:

• Q&A with grounding (RAG-style experiences)
• summarization of long content (cases, conversations, documents)
• classification (routing, tagging, prioritization)
• tool-using assistants that call endpoints to retrieve or act

# What not to use LLMs for

Avoid using an LLM as the only source of truth for:

• permission checks
• deterministic business rules
• data updates that require strict correctness

Instead, move those into endpoints and validate outputs.

# A recommended architecture pattern

• Search/Graph: retrieve candidates and context
• Endpoints: implement business logic and tool calls
• LLM: synthesize the user-facing response
• UI: present results with navigation and traceability (what sources were used?)

# Next steps

• Configure providers and templates: LLM Configuration
• Learn practical prompting patterns: Prompting Patterns