AI Glossary
The language of
AI governance.
A practical reference for agents, data protection, MCP, routing, budgets, verification, and AI Signals.
AI foundations
- Agent
- A system that uses a model to take multi-step actions, such as calling tools, running code, or making requests, toward a goal, rather than just producing a single text response.
- Context window
- The maximum amount of text, measured in tokens, a model can consider at once: the prompt, conversation history, and its own response, combined.
- Embedding
- A numerical representation of text, or other data, that captures meaning. Embeddings let systems compare or search content by similarity instead of exact keyword matches.
- Fine-tuning
- Additional training on a pretrained model using a narrower dataset, to adapt its behavior for a specific task or domain.
- Foundation model
- A broadly trained model that can support many tasks and be adapted for particular applications.
- Hallucination
- When a model produces an answer that sounds plausible but is factually wrong or unsupported by its sources.
- Large language model (LLM)
- A model trained on large amounts of text to understand and generate natural language. It's the core technology behind chatbots, coding assistants, and most modern AI agents.
- Local model
- A model hosted within your own environment rather than accessed through an external provider endpoint.
- Model Context Protocol (MCP)
- An open standard that lets AI models and agents call external tools and data sources in a consistent way, instead of every integration needing custom code.
- Multimodal workload
- A task that uses more than one kind of content, such as text, images, audio, video, or files.
- Open-weight model
- A model whose trained weights are available for use under its license. Open weights do not necessarily mean the training data, code, or all uses are unrestricted.
- Prompt
- The input sent to a model, including instructions, questions, or context, that shapes what it responds with.
- Request
- The call an application, agent, or user sends to a model provider: the prompt, plus parameters like which model to use and authentication. It's the point where a gateway can inspect and control traffic before it ever reaches the provider.
- Response
- What the model provider sends back after processing a request, usually generated text, but it can also include tool calls, token usage, or an error. Guardrails can check a response just as they check the request that triggered it.
- Retrieval-augmented generation (RAG)
- A technique that retrieves relevant information at request time and supplies it as context for a model. Retrieval can improve grounding, but an answer still needs to be checked against the sources.
- Temperature
- A setting that controls how random or deterministic a model's output is. Lower values produce more predictable answers; higher values produce more varied ones.
- Token
- The basic unit a model reads and generates text in, roughly a word or part of a word. Usage-based AI pricing is usually billed per token.
- Tool calling (function calling)
- A model capability where the model can request that a specific tool, API, or function be run, then use the result to continue its task.
Governance & data protection
- AI Border Gateway (AIBG)
- Cortega’s gateway for governing model and MCP traffic, with routing, policy controls, observability, and centralized access management.
- AI Firewall
- A control layer for AI interactions that inspects content, permissions, and actions. In Cortega, this solution is powered by Cortega Guardrails.
- AI gateway
- Infrastructure that sits between AI callers (apps, agents, employees) and model providers, so requests can be inspected, controlled, and logged in one place instead of scattered across every integration.
- AI governance
- The practices and controls that let an organization know where AI is used, apply policy to that use, and produce evidence of what happened, across every model, agent, and tool involved.
- AI TRiSM
- Short for AI Trust, Risk, and Security Management, a category (coined by Gartner) covering the practical work of governing AI: visibility, runtime inspection, policy enforcement, data protection, access control, and audit evidence.
- Audit evidence
- The recorded detail, such as identity, policy decision, data involved, model used, and outcome, that lets a compliance or security reviewer verify what actually happened during an AI interaction.
- Budget cap
- A spend limit enforced on AI usage, typically scoped to a team or provider, so a single integration or feature can't run away with cost.
- Data exfiltration
- Unauthorized movement or disclosure of data outside its permitted destination or access boundary.
- Endpoint Guard
- Cortega’s product for governing employee AI activity on computers and applying endpoint policies to supported workflows.
- Failover
- Automatically shifting traffic to a backup model or provider when the primary one is slow, unavailable, or returning errors.
- Guardrails
- Automated checks applied to AI requests and responses, such as detecting sensitive data or blocking a request, enforced before or during a call rather than reviewed after the fact.
- Model routing
- Directing an AI request to a specific model or provider based on rules like cost, availability, or task type, instead of hardcoding one model per application.
- Prompt injection
- An attack where malicious instructions are hidden in content a model processes, such as a document, a web page, or a tool response, to make the model act against its intended instructions.
- Redaction
- Removing or masking sensitive portions of content before that content is passed on or displayed.
- Role-based access control (RBAC)
- A permissions model where what a user or system can do is determined by the role they're assigned, rather than being configured person by person.
- Shadow AI
- AI tools, agents, or API keys in use across an organization without IT or security's knowledge, often set up directly by individual employees or teams.
- Tenant isolation
- Keeping one customer’s or organizational tenant’s data, access, policies, and budgets separate from another’s.
- Threat level
- A representation of detected risk that can affect the caution and controls applied to subsequent activity.
- Virtual key
- A credential issued by a governance layer, instead of a raw provider API key, that lets a team or application call AI providers without ever holding the underlying key directly.
- Zero trust
- A security approach that verifies every request based on identity and context, rather than assuming anything inside a network perimeter is automatically trusted.
Agents, MCP & tools
- Agent poisoning
- Manipulating content an agent consumes so that it changes its behavior or takes an unauthorized action.
- Agent trajectory
- The sequence of steps and tool choices an agent takes while pursuing a task. Analysis can identify loops, wrong-tool use, or abandonment.
- Application-specific inspection
- Checking the meaning and structure of a particular application’s records, beyond generic content checks or permission to call a tool.
- MCP server
- A service that exposes tools, resources, or prompts through the Model Context Protocol for compatible clients.
- Model interworking
- Using multiple models within an AI environment or workflow, with different models serving different tasks or stages.
- Production agent
- An AI agent operating on live workflows, users, or data, with requirements for access control, reliability, observability, and cost management.
- vCon record
- A structured conversation record. Cortega’s vCon example inspects user consent and law enforcement indicators in records accessed through MCP.
Routing & resilience
- 5xx error
- An HTTP server-side error. In model access, it can indicate that a provider or upstream service could not complete the request.
- Centralized key management
- Managing provider credentials through a shared control layer instead of distributing raw keys across applications and agents.
- Fault tolerance
- The ability to continue operating when a component fails, using mechanisms such as approved fallbacks and redundant capacity.
- Latency-based routing
- Selecting among eligible model destinations based on response latency.
- Least-cost routing
- Choosing the lowest-cost eligible model within the workload’s policy and capability constraints.
- Load balancing
- Distributing requests across eligible model endpoints to share demand and capacity.
- Model evaluation
- Assessing model behavior on defined tasks and criteria, such as quality, safety, cost, and latency.
- Model study
- A structured comparison of models using a chosen workload and evaluation criteria to inform selection or switching.
- Rate limit
- A provider’s allowance for requests or tokens over a period. Exceeding it can cause requests to be delayed or rejected.
- Semantic routing
- Selecting a model using the meaning or intent of a request, rather than only a fixed endpoint or model name.
Budgets & efficiency
- Agent budget
- A usage or spending allowance assigned to an individual agent or a team of agents.
- Budget-aware routing
- Using available allowance and department policy as inputs to model selection.
- Graceful budget degradation
- Adapting model cost or permitted usage as an allowance runs low, within policy and available alternatives.
- Model allowlist
- The set of models a particular agent, identity, or department is permitted to use.
Verification & AI Signals
- Aegis
- A Cortega Agent with an interactive team chat interface and a traffic anomaly watcher that raises notifications.
- AI Signals
- Findings from analysis of AI activity, such as sentiment, policy compliance, cost efficiency, workload categories, or agent behavior.
- AI Verifier
- Cortega’s response verification product for inline policy, legal citation, factual relevance, and contextual accuracy checks. Its primary workflow requires no plugins, file uploads, or MCP server.
- Citation verification
- Checking whether a cited source exists and whether its details and support for a claim are valid. A real source can still be cited incorrectly.
- Contextual accuracy
- How accurately a response uses and reflects the context available in the interaction.
- Cortega Agents
- Independently enabled analysis agents that classify, score, or watch opted-in recorded traffic from AI Border Gateway and Endpoint Guard.
- Factual relevancy
- How well a response’s factual content addresses the question or task.
- Red teaming
- Testing an AI system with challenging or adversarial inputs to evaluate its behavior and controls.
- Rubric
- A defined set of criteria used to score a conversation or assess compliance with a policy or brand standard.
- Sentiment & escalation
- Signals of emotion or frustration in a conversation that can help identify interactions needing review.
- Workload classification
- Assigning an exchange to a task category for reporting and model-fit analysis.
No matching terms. Try a different term or clear the search.
MAKE YOUR NEXT MOVE
Book a demo
Explore how Cortega fits your AI environment.