Home › Glossary

AI glossary

Plain-language definitions of 236 terms you will meet when choosing and using AI tools.

ABM (account-based marketing): A go-to-market approach that targets named accounts with coordinated sales and marketing; AI platforms often score account intent.
Active learning: A labeling approach in which the model picks the examples it is least sure about so people label the most useful data first.
Adaptive machine translation: Machine translation that adjusts its output using a customer's corrections and translation memories as it is used.
Adoption agent: An AI that watches product usage and nudges customers toward features that improve retention.
Adversarial AI testing: Using AI to attack or stress-test another AI system to find failures, jailbreaks or security gaps.
Agent (AI agent): A program that uses a language model to decide on steps, call tools or APIs, and work toward a goal with limited step-by-step instruction from a person.
Agent assist: AI that listens to or reads a live customer conversation and suggests answers, knowledge articles or next steps to the human agent.
Agent billing: Metering and charging for AI agent usage, such as runs, tool calls or compute time, often for AI-native products.
Agent memory: Stored information that lets an AI agent recall facts, preferences or past steps across sessions, often kept in a vector store, knowledge graph or database outside the model.
Agent sandbox: An isolated computer or container where an AI agent can run code and use tools without affecting the host system.
Agent SDK: A software development kit that provides the building blocks for AI agents, such as the model loop, tool calling, memory and tracing.
Agentic IDE: A code editor in which AI agents can plan multi-step tasks, edit files, run commands and check results, with the developer reviewing the work.
Agentic workflow: A process in which a model plans, takes actions, checks results and repeats, instead of answering a single prompt once.
AI avatar video: Video in which a generated or cloned presenter speaks a script, used for training, marketing and sales outreach.
AI bookkeeping: Automating bookkeeping tasks such as categorizing transactions, reconciling accounts and preparing books with AI assistance.
AI companion: A chat app built around a persistent character meant for ongoing personal conversation rather than task completion.
AI content detector: Software that estimates whether text was written by an AI model; scores are probabilistic and can be wrong.
AI dubbing: Replacing the speech in a video with speech in another language using AI translation and voice generation, sometimes keeping the original speaker's voice.
AI dubbing: Translating a video's speech into another language with AI voices, often keeping the original speaker's voice and timing.
AI gateway: A service that sits between an application and one or more model providers, offering a single API plus features such as fallbacks, caching, logging and spend limits.
AI humanizer: A tool that rewrites AI-generated text to read more like human writing; results vary and may conflict with school or publisher rules.
AI motion capture: Estimating body movement from ordinary video with AI and turning it into 3D animation data, without suits or markers.
AI search visibility: How often and how accurately a brand, product or page is mentioned in answers from AI assistants and AI search features.
AI security posture management (AI-SPM): Discovering the AI models, agents and apps an organization uses and checking them for misconfigurations, risky permissions and data exposure.
AI SOC analyst: Software that uses AI to investigate security alerts, gather evidence and suggest whether an alert is a real threat.
AI takeoff: Automated measurement of quantities from construction drawings or plans, often for bidding.
AI website builder: A tool that generates a website or landing page from a short description, often including copy, layout and basic hosting.
Ambient clinical documentation: AI that listens to a clinician-patient visit and drafts the clinical note for the clinician to review and sign.
Ambient scribe: An AI system that listens to a clinical visit and drafts notes or documentation for a clinician to review.
Answer engine optimization (AEO): Work aimed at getting a brand or page mentioned or cited in answers from AI assistants and AI search features. Closely related to GEO.
API (application programming interface): A defined way for one piece of software to request data or actions from another. Model providers expose APIs so developers can send prompts and receive responses.
Applicant tracking system (ATS): Software that stores job postings, applications and candidate records and moves candidates through hiring stages. Many AI recruiting tools connect to one.
ATS resume checker: A tool that compares a resume with a job description and flags formatting or keyword gaps that may affect applicant tracking system screening.
Automated employment decision tool (AEDT): Software that scores or ranks job candidates or employees; some places, such as New York City, require bias audits and notices before using one.
Automated resolution: A customer conversation fully resolved by an AI agent without a human; many support tools bill AI usage per automated resolution.
Autonomous penetration testing: Using AI agents to probe applications and networks for vulnerabilities and attempt to exploit them, the way a human penetration tester would.
Autonomous pentesting: Software that safely runs attack techniques against your own systems to find exploitable weaknesses without a human tester for each step.
Background removal: Automatically separating the subject of a photo from its background so the background can be made transparent or replaced.
Behavior analytics: Tools that study how users interact with websites or apps, using recordings, heatmaps and funnels rather than only survey answers.
Benchmark: A standard test set used to compare models on a specific skill, such as coding, math or reading comprehension. Scores depend on the test and how it is run.
Brand mention monitoring: Tracking when a brand name appears in AI answers, search results or social content so teams can measure visibility and respond.
Chain of thought: A prompting style or model behavior in which the model writes out intermediate reasoning steps before giving a final answer.
Chatbot: A program that holds a conversation with a user in text or voice. Modern chatbots are usually built on large language models.
Clinical documentation: Notes and records of a patient encounter. AI scribes draft this documentation from audio or transcripts for clinician review.
Cloud lab: A remote laboratory environment that an AI research agent can use to run experiments or analyses.
Coding harness: A wrapper around coding agents that standardizes how they read repos, run commands and report results.
Compliance automation: Software that collects security evidence and maps it to frameworks such as SOC 2 or ISO 27001 so audits take less manual work.
Computer use: An AI capability in which a model operates a computer or browser by reading the screen and issuing clicks and keystrokes.
Computer-assisted translation (CAT) tool: An editor for professional translators that splits text into segments and suggests matches from translation memory and machine translation.
Computer-use agent: An AI agent that operates a computer or browser by reading the screen and issuing clicks and keystrokes to complete tasks.
Contact center as a service (CCaaS): Cloud software that runs a contact center, including call routing, queues, recording and reporting, without on-premise phone hardware.
Context engineering: Choosing and arranging the instructions, documents, tool results and memory that go into a model's context window so it can do a task well.
Context window: The maximum amount of text, measured in tokens, that a model can consider at one time, including the prompt and its own reply.
Continuous learning platform: Software that keeps improving models or agents from new data and feedback after the initial training.
Contract AI: Software that uses language models to review, extract clauses from or draft legal contracts, usually with a lawyer in the loop.
Contract lifecycle management (CLM): Software that manages contracts from request and drafting through approval, signature, obligations and renewal.
Conversational recruiting: Using a chat or text assistant to answer candidate questions, screen applicants and schedule interviews.
Copilot: A common product name for an assistant built into another tool, such as an editor or office suite, that suggests or completes work alongside the user.
Customer satisfaction score (CSAT): A support metric based on asking customers to rate an interaction, often used to compare AI and human-handled conversations.
Data labeling (annotation): Adding tags, categories or other labels to raw data such as images, text or audio so it can be used to train or evaluate models.
Data poisoning: An attack that inserts misleading or malicious examples into training data so a model learns harmful behavior.
Data security posture management (DSPM): Finding and classifying sensitive data across cloud and SaaS systems and flagging who can access it and where it is exposed.
Deep learning: A branch of machine learning that trains multi-layer neural networks on large amounts of data.
Deepfake detection: Tools that analyze audio, images or video for signs that they were generated or manipulated by AI.
Device cloud: A service that rents physical phones or computers so AI agents can operate real devices over the network.
Differential diagnosis assistant: Software that suggests possible conditions from symptoms and findings so a clinician can consider and rule them out.
Diffusion language model: A language model that produces text by refining a whole draft over several steps, rather than writing one token at a time.
Diffusion model: A type of model, widely used for images and video, that learns to turn random noise into a coherent picture step by step, guided by a text prompt.
Digital pathology: Scanning tissue slides into high-resolution images so they can be reviewed on screen and analyzed by software, including AI.
Distillation: Training a smaller model to imitate the outputs of a larger one, so it keeps much of the larger model's ability at lower cost.
Distributed GPU network: A marketplace that pools GPUs from many providers for AI training or inference jobs.
Document intelligence API: An API that parses PDFs and images into structured text, tables and layout for search, RAG or downstream automation.
Document parsing: Turning files such as PDFs, scans and spreadsheets into structured text and data, including tables and layout, so software or models can use them.
Embeddable badge: A small image and link that a website can place on its pages to show it is listed in a directory.
Embedding: A list of numbers that represents the meaning of a piece of text, image or other data so that similar items sit close together and can be searched by similarity.
Evaluation (evals): Tests that measure how well a model or an AI application performs on tasks that matter to its builder, often run automatically before each release.
Experiment tracking: Recording the code, data, settings and results of each model training run so runs can be compared and reproduced.
Few-shot prompting: Including a small number of worked examples in a prompt so the model can follow the same pattern.
Fine-tuning: Continuing to train an existing model on a smaller, specialized dataset so it performs better on a particular task or style.
Focus time: Blocks on a calendar reserved for deep work; AI calendar tools often protect or auto-schedule these around meetings.
Foundation model: A large model trained on broad data that can be adapted to many tasks, such as a general-purpose language or image model.
FP&A: Financial planning and analysis: budgeting, forecasting and management reporting that helps leaders plan revenue and spend.
Function calling (tool use): A feature that lets a model return a structured request to run a named function or tool, which the application then executes and reports back.
Gaussian splatting: A way of representing and rendering 3D scenes as many small, soft, colored blobs, often built from photos or video of a real place.
Gaussian splatting: A 3D capture technique that represents a scene as many small, colored blobs, giving realistic real-time views from photos or video.
Generative AI: Models that create new content, such as text, images, audio, video or code, based on patterns learned from training data.
Generative engine optimization (GEO): Work aimed at making a brand or page show up accurately in answers from AI assistants and AI search engines. Also called answer engine optimization (AEO).
GGUF: A file format widely used to distribute quantized language models for local runners such as llama.cpp and KoboldCPP.
GPU: A graphics processing unit, a chip that performs many calculations in parallel. GPUs are the main hardware used to train and run large models.
GPU marketplace: A service that lets people rent GPUs from many independent providers, usually priced by the hour.
Graph database: A database that stores data as nodes and relationships, useful for knowledge graphs and GraphRAG retrieval.
GraphRAG: Retrieval-augmented generation that uses a knowledge graph of entities and relationships, not only text chunks, to find context for a model.
Ground truth: Correct answers or labels used as the reference when training or evaluating a model.
Grounding: Supplying a model with trusted source material, such as documents or search results, so its answer is based on that material.
Guardrails: Rules, filters or checks placed around a model to limit unwanted inputs or outputs, such as unsafe content or off-topic answers.
Hallucination: When a model states something that sounds plausible but is false or unsupported by its sources.
Harness as a Service: A managed cloud runtime that runs AI agents with sandboxes, tools and session controls so teams do not operate the agent infrastructure themselves.
Headless browser: A web browser that runs without a visible window and is controlled by code, often used by automation scripts and browser agents.
Heatmap: A visual overlay showing where users click or move on a page, used to spot ignored buttons or confusing layouts.
Help desk: Software that gathers customer requests from email, chat, social and other channels into tickets that support agents can assign, track and resolve.
Hybrid search: Search that combines keyword matching with vector (semantic) similarity, so results match both exact terms and overall meaning.
IDE extension: A plugin that adds features such as AI code completion or chat inside an existing code editor.
Image vectorization: Converting a pixel image such as a PNG or JPG into vector shapes (for example SVG) that scale without losing sharpness.
Image-to-3D: Generating a 3D model from one or more photos or drawings of an object.
Inference: Running a trained model to produce an output. Inference cost and speed are key factors when choosing a model provider.
Inference endpoint: A hosted URL where an application sends inputs to a deployed model and receives predictions, usually billed by usage.
Intelligent document processing (IDP): Using OCR and AI models to extract fields and tables from invoices, forms and other documents and route them into business systems.
Intelligent document processing (IDP): Automation that classifies documents and extracts their data with OCR and AI, then sends it to business systems.
Interactive voice response (IVR): A phone menu system that routes callers by keypad or voice; AI voice agents increasingly replace fixed IVR menus with natural conversation.
Interview intelligence: Recording and transcribing job interviews so that notes, scorecards and summaries can be produced automatically.
iPaaS: Integration platform as a service: a cloud tool for connecting apps and moving data between them.
Jailbreak: A prompt crafted to make a model ignore its safety rules or intended instructions.
Knowledge graph: A way of storing information as entities and the relationships between them, which AI systems can query to find connected facts.
Large language model (LLM): A neural network trained on large amounts of text to predict and generate language. LLMs power most chatbots and writing assistants.
Latency: The delay between sending a request and receiving a response. Voice and real-time applications are especially sensitive to it.
Lip sync: AI that edits mouth movements in a video so they match new or translated speech.
LLM gateway: A proxy in front of one or more model APIs that adds a shared interface, logging, rate limits, fallbacks and sometimes routing.
LLM serving framework: Software such as an inference server that loads a model onto GPUs and handles batching, caching and streaming of requests efficiently.
LLM-as-a-judge: Using a language model to grade another model's outputs against criteria such as accuracy or tone, as part of an evaluation.
Local LLM frontend: A desktop or web UI that talks to language models running on your own machine or a private server instead of only a hosted API.
Localization (l10n): Adapting a product or content for a specific language and region, including translation, formats, currency and cultural details.
Localization platform: Software that manages translation of apps, websites and docs across languages, often with translation memory and AI suggestions.
Long-to-short clipping: Automatically turning a long video into short clips for social feeds, often with captions and hooks.
LoRA: Low-Rank Adaptation, a technique that fine-tunes a model by training a small set of added weights instead of the whole model, which lowers cost.
Machine learning: Building software that improves at a task by learning patterns from data instead of following only hand-written rules.
Machine translation: Automatic translation of text or speech from one language to another by software, today usually by neural networks or language models.
Machine translation post-editing (MTPE): A workflow where a human translator reviews and corrects machine translation output instead of translating from scratch.
Mastering (audio): The final step of music production that balances loudness, tone and dynamics so a track sounds consistent across playback systems.
MCP (Model Context Protocol): An open protocol for connecting AI applications to external tools and data sources through a common interface, so one integration can work across different clients.
MCP gateway: A service that connects AI agents to many tools through the Model Context Protocol with authentication and policy controls.
Mixture of experts (MoE): A model design in which only some specialized sub-networks are used for each input, which can reduce the computing needed per response.
MLOps: Practices and tools for building, deploying, monitoring and maintaining machine learning models in production.
Model: The trained system, made of numerical weights, that takes an input and produces an output such as text, an image or a prediction.
Model router: Software that sends each request to one of several models based on rules or a learned policy, often to balance cost and quality.
Model routing: Automatically choosing which model handles each request, for example sending simple prompts to a cheaper model and harder ones to a stronger model.
Model scanning: Inspecting model files and dependencies for malicious code, backdoors or unsafe serialization before they are used.
Model serving: Running a trained model behind an API or service so applications can send it requests.
Model serving: Running trained models behind an API so applications can send inputs and get predictions at scale.
Multi-model assistant: A chat product that lets the user switch between or combine several language or image models in one interface.
Multimodal: Able to work with more than one kind of input or output, such as text, images, audio and video.
Multiplayer AI office: A shared workspace where people and AI agents work side by side on tasks and documents.
Network detection and response (NDR): Security tools that watch network traffic to find suspicious behavior and help respond to it.
Neural network: A computing system made of layers of connected units whose strengths are adjusted during training.
Neural TTS: Text-to-speech systems that use neural networks to produce more natural-sounding speech than older concatenative or parametric methods.
Observability: Tools and practices for recording and inspecting what an AI application did, including prompts, responses, costs and errors, so problems can be traced.
OCR (optical character recognition): Technology that reads printed or handwritten text from images and scanned documents and converts it into machine-readable text.
Omnichannel inbox: A single workspace where support or sales teams handle conversations from email, live chat, WhatsApp, social media and other channels.
On-call AI engineer: An AI system that triages production alerts, investigates causes and can open fix pull requests.
Open source: Software released under a license that lets others view, use, change and share the source code, within the license terms.
Open-weight model: A model whose trained weights are published so others can download and run it. Licenses vary and may not be open source in the strict sense.
OpenTelemetry: An open standard and set of SDKs for collecting traces, metrics and logs; many LLM observability tools use it to record model and tool calls.
Orchestration: Coordinating the steps of an AI application, such as calling models, tools and databases in order, often with a framework.
Parallel dialer: Sales software that calls several numbers at once and connects the rep only when someone answers.
Parameters: The adjustable numerical values inside a model that are set during training. Parameter count is one rough indicator of model size.
People intelligence: Software that analyzes workforce or people data to help with hiring, retention or organization design.
Permission prompt relay: A system that forwards an AI agent's permission requests to a human on another device and records who approved them.
Personal AI pendant: A wearable microphone device that records conversations and feeds an AI assistant that builds searchable memory for the wearer.
Photogrammetry: Building a 3D model from many overlapping photos of an object or space.
Playbook (legal): A set of preferred and fallback clause positions a legal team uses when reviewing or negotiating contracts, often encoded for AI review.
Presentation generator: A tool that drafts a full slide deck, including layout, text and images, from a prompt, outline or document.
Product analytics: Measuring how people use a product with events, funnels and retention charts so teams can see what drives engagement or churn.
Product observability: Tracking how a product behaves in production, including AI features, so teams can find regressions and quality issues.
Prompt: The input given to a model, which can include instructions, questions, examples and background material.
Prompt caching: Reusing the processed form of a repeated prompt prefix across requests, which can lower cost and latency for long, shared context.
Prompt engineering: Writing and refining prompts to get more useful, reliable output from a model.
Prompt injection: An attack in which text in a document, web page or message tries to override the instructions given to an AI application.
Quantization: Storing a model's numbers at lower precision to cut memory use and speed up inference, sometimes with a small loss in quality.
Radiology AI: Software that analyzes medical images such as X-rays or CT scans to highlight findings for radiologists.
RAG (retrieval-augmented generation): A pattern in which an application first retrieves relevant documents, often from a vector database, and gives them to the model so its answer draws on that material.
Rate limit: A cap set by a service on how many requests or tokens you can use in a given period, such as per minute or per day.
Reasoning model: A model trained or configured to spend extra computation working through a problem in steps before it answers.
Red teaming: Deliberately probing an AI system with adversarial inputs to find harmful, unsafe or incorrect behavior before users do.
Reinforcement learning from human feedback (RLHF): A training method that uses people's ratings of model outputs to steer the model toward preferred behavior.
Remote simultaneous interpretation (RSI): Live interpretation of a meeting or event delivered over the internet, by human interpreters or AI speech translation.
Reranking: A second step in search or RAG where a model reorders an initial list of results so the most relevant ones come first.
Retopology: Rebuilding a 3D model's mesh with a cleaner, lighter set of polygons so it is easier to animate, texture or use in games.
Revenue intelligence: Software that captures sales calls, emails and CRM activity and uses AI to forecast deals, flag risks and coach reps.
RPA (robotic process automation): Software bots that repeat rule-based tasks in other applications, such as copying data between systems.
Rubric-based grading: Scoring student work against defined criteria; AI grading assistants use a teacher's rubric to draft scores and feedback for the teacher to review.
Sales enablement: Content, training and coaching tools that help sales teams prepare for and run customer conversations.
Sales engagement platform: Software for sequencing emails, calls and other outreach to prospects, often with AI help for writing and timing.
Sales roleplay: Practicing sales calls with an AI that plays the buyer, then getting scored feedback on the conversation.
Sandbox: An isolated environment where code or an agent can run without being able to affect the rest of a system.
Scheduling link: A personal or team URL that shows available times and lets others book a meeting without email back-and-forth.
Search API: An API that returns web search results as structured data for apps and AI agents.
Secrets scanning: Searching code and repositories for exposed credentials such as API keys and passwords.
Security information and event management (SIEM): A system that collects and analyzes security logs from across an organization to detect and investigate threats.
Security operations center (SOC): The team, often with dedicated tools, that monitors an organization's systems and responds to security alerts.
Security orchestration, automation and response (SOAR): Tools that automate security workflows such as enriching alerts, isolating devices and opening tickets, often across many security products.
Self-improving memory: A memory layer that updates what an agent remembers based on outcomes, so later tasks can use what worked before.
Semantic search: Search that matches the meaning of a query rather than only its exact words, usually by comparing embeddings.
Serverless GPU: A cloud service that runs AI workloads on GPUs on demand and bills for the compute used, without the customer managing servers.
Session replay: A recording of a user's clicks, scrolls and page views that can be played back to debug issues; often paired with heatmaps.
Shadow AI: Use of AI tools inside an organization without approval or oversight from IT or security teams.
Singing voice synthesis: Generating sung vocals from lyrics and melody, or converting one singer's voice to another, with AI models.
Small language model (SLM): A language model with relatively few parameters, designed to run cheaply or on local devices such as phones and laptops.
SOC 2: A common security audit report for service companies, based on trust service criteria such as security, availability and confidentiality.
Social listening: Monitoring public posts, news and forums for mentions of a brand or topic to measure sentiment and spot trends.
Social media scheduler: Software that queues posts to publish on social networks at set times, often with AI help for drafting and repurposing content.
Software composition analysis (SCA): Scanning the open-source packages an application depends on for known vulnerabilities, license issues or malicious code.
Software supply chain attack: An attack that compromises a dependency, build tool or package registry so malicious code reaches the software that uses it.
Speaker diarization: Labeling which speaker said each part of a recording, so a transcript shows who spoke when.
Speech-to-text (STT): Software that converts spoken audio into written text, also called transcription or automatic speech recognition.
Spend management: Software for company cards, expenses, bills and procurement in one place, often with automated policy checks.
Static application security testing (SAST): Analyzing source code without running it to find security flaws such as injection or unsafe data handling.
Stem separation: Splitting a mixed song into parts such as vocals, drums and bass, often with AI.
Structured output: Model output constrained to a fixed format such as JSON that matches a schema, so software can read it reliably.
Subscription analytics: Tracking SaaS metrics such as MRR, churn and customer lifetime value, often with cohort charts and revenue recognition views.
Synthetic data: Artificially generated data that mimics the patterns of real data, used for training, testing or sharing without exposing real records.
Synthetic data platform: A service that generates artificial datasets that preserve statistical patterns of real data while reducing privacy risk.
System prompt: Instructions given to a model by the application developer, separate from the user's message, that set its role and rules.
Talent intelligence: Using data and AI to map skills across job seekers and employees, for example to match candidates to roles or suggest internal moves.
Temperature: A setting that controls how varied a model's output is. Lower values give more predictable text and higher values give more varied text.
Text-to-3D: Generating a 3D model, with shape and often textures, from a written description.
Text-to-CAD: Generating an editable computer-aided design (CAD) model of a part from a written description, rather than a mesh used only for display.
Text-to-speech (TTS): Software that turns written text into spoken audio, often with a choice of voices.
Text-to-video model: An AI model that generates video clips from a written prompt, and often from a starting image, without filming.
Therapy note assistant: An AI tool that drafts progress notes from a therapy session for the clinician to review and edit.
Time to first token (TTFT): How long a model takes to start returning its response after receiving a request. An important speed measure for chat and voice apps.
Token: A small unit of text, such as a word or part of a word, that a model reads and writes. Many APIs bill by the number of tokens.
Tracing: Recording each step of an AI application's run, such as prompts, model calls and tool calls, so developers can debug and measure it.
Training data: The examples a model learns from. Its content and quality strongly affect what the model can do.
Transformer: A neural network design built around attention, which lets a model weigh the relationship between all parts of its input. Most modern language models use it.
Translation management system (TMS): Software that organizes translation projects, translation memories, glossaries and reviewers, often with machine translation built in.
Translation memory: A database of previously translated sentences that translation tools reuse to keep wording consistent and avoid translating the same text twice.
UGC ad generator: Software that creates user-generated-style video ads with AI avatars or scripts for social platforms.
Vector database: A database designed to store embeddings and quickly find the ones most similar to a query.
Vibe coding: Building software mainly by describing what you want to an AI tool in plain language and accepting or adjusting the code it writes, often with little manual coding.
Virtual staging: Adding furniture and decor to photos of empty rooms with software, often used in real estate listings.
Visitor identification: Matching anonymous website traffic to companies or people so sales teams can follow up on high-intent visitors.
Voice agent: An AI system that holds a spoken phone or voice conversation, understanding speech and responding with generated audio.
Voice cloning: Creating a synthetic voice that imitates a particular person's voice from recordings, which raises consent and misuse concerns.
Voice harness: A developer framework for building voice AI applications, including speech input, model calls and tool actions.
Web scraping: Automatically extracting content or data from web pages, for example to feed search indexes, datasets or AI agents.
Website visitor identification: Matching anonymous website visits to companies or individuals so sales teams can follow up; subject to privacy laws in many regions.
Zero-shot prompting: Asking a model to perform a task with instructions only, without giving examples.