Glossary of Terms

Context Window:
​Definition: The maximum number of tokens (text units) a large language model (LLM) can process in a single interaction.
​
​Example: A 4,000-token context window allows reviewing a 3,000-word document in one go.
​

Retrieval-Augmented Generation (RAG):
​Definition: A technique where AI models retrieve external information (like documents or databases) to enhance responses, reducing reliance on pre-trained knowledge.
​
​Example: An agent references an uploaded RFP to answer deadline questions.
​

Temperature:
​Definition: A parameter (0–1) controlling the randomness or creativity of AI outputs.
​
​Example: Low temperature for grading consistency; high temperature for brainstorming.
​

Tokens:
​Definition: Units of text (words, subwords, or characters) processed by AI models.
​Relevance to Rolai: Impacts how agents handle document analysis and workflow automation.
​
​Example: “Rolai” = 1 token; “university” = 2 tokens.
​

Hallucination:
​Definition: When an AI generates incorrect or fabricated information not grounded in its training data or provided context.
​
​

Agent:
​Definition: A customizable AI assistant in Rolai designed for specific roles (e.g., grading, research, compliance). Agents combine models, instructions, guardrails, and integrations.
​
​Example: A “Grading Assistant” agent analyzes student submissions using a rubric.
​

Flow:
​Definition: A visual, no-code automation that connects triggers, actions, and AI capabilities to streamline multi-step processes.
​
​Example: Trigger daily at 8 AM → run research agent → email summaries.
​

Trigger:
​Definition: An event or condition that starts a Flow.
​
​Types: Manual (user-initiated), Scheduled (runs at set times), App-Based (external event).
​
​Example: A Flow triggered by a new file upload in OneDrive.
​

Action:
​Definition: A step within a Flow that performs a task, such as running code, sending an email, or invoking an AI Agent.
​

Control Flow:
​Definition: Logic elements in Flows (like If...Else or Loop) that allow for branching, decision-making, or repeating steps.
​

Knowledge:
​Definition: A secure repository for organizational documents (PDFs, txt, docx, doc, pptx, xlsx, etc.) that agents reference via RAG.
​

Prompts (Formerly known as Smart Tasks):
​Definition: A pre-built, one-click AI workflow for common, repeatable tasks (e.g., summarization, data extraction).

Spaces:

A dedicated area where users can manage specific settings, agents, and flows, ensuring that workflows can be shared selectively while maintaining privacy.
​

Integration Framework / Integration:
​Definition: Rolai’s system for connecting external tools (like Canvas, Microsoft 365, Google Workspace, HubSpot, etc.) to agents and Flows, enabling seamless data exchange and automation.
​
​Example: A Flow that saves agent outputs to Google Drive or posts grades to Canvas.
​

LLM (Large Language Model):
​Definition: An advanced AI model trained on large datasets to understand, generate, and process human language.
​
​Examples: OpenAI GPT-4, Google Gemini, Anthropic Claude-all accessible in Rolai.
​

Chat (AI Workbench):
​Definition: Rolai’s unified workspace for accessing, experimenting with, and managing leading AI models for research, content creation, and analysis.
​

Compliance:
​Definition: Adherence to regulations (GDPR, FERPA) and institutional policies.
​

Best Practices for Usage:

Support

If you need help or clarification on any term or feature, reach out via the Intercom chat or email support@rolai.com.