AI glossary

Plain-English definitions for the terms people keep hearing.

You don't need to memorize these. You only need enough understanding to make better decisions when using AI.

Search glossary

Basic AI terms

AI: Software that can perform tasks usually requiring human-like reasoning, language, pattern recognition, or decision support.
Generative AI: AI that creates something new, such as text, images, summaries, plans, or code.
Model: The system behind the AI. A model is trained on patterns and uses them to respond to requests.
Prompt: The instruction or question you give an AI system.
Output: The result the AI gives back.
Context: The information the AI has available while answering or working on a task.
Training: The process of building a model by exposing it to large amounts of information.
Inference: The moment a trained model is used to produce an answer or result.
Hallucination: When AI gives an answer that sounds confident but is wrong, unsupported, or invented.
Grounding: Giving AI specific information to work from, so its answer is tied more closely to real sources or documents.

AI agent: An AI system that can take steps toward a goal, not just answer a single question.
Workflow: A repeatable set of steps for getting work done.
Tool: Something an agent can use to complete a task, such as reading a file, checking a calendar, searching information, or preparing a draft.
Action: Something the agent does as part of a workflow.
Approval: A point where you must say yes before the agent continues.
Human in the loop: A design where the person stays involved in important decisions instead of handing everything to the AI.
Autonomy: How much the agent can do on its own before it needs you.
Record: A clear history of what happened during a workflow.
Permission: What the agent is allowed to see or do.
Guardrail: A limit that keeps an AI system from doing something unsafe, unwanted, or outside its purpose.

Local-first: A design where work happens on your own computer whenever possible.
Local model: An AI model that runs on your own machine.
Cloud model: An AI model that runs on outside servers.
Private data: Information that should be handled carefully, such as business documents, client information, health information, financial details, credentials, or personal messages.
Token: A small piece of text an AI model reads or produces. AI services often measure usage by tokens.
Token cost: The cost created when an AI service charges based on how much text is processed or produced.
Token budget: A limit you set on how much an AI task is allowed to spend before it runs.
Model routing: Choosing which model handles a task: your free local model, or a paid cloud model. In Automatic, OOMU keeps the work on your Mac and uses a cloud model only when you have allowed it and the request calls for one.
Local pre-processing: Preparing and trimming a task on your own Mac before sending anything to the cloud, so the cloud request is smaller, cheaper, and more private.
Reasoning level: How much effort a model spends thinking before it answers. Lower levels are faster and cheaper; higher levels deliberate more.
Encryption: A way of protecting information so it can't be read without the right key.

Index: A prepared map of information that helps software search files or documents quickly.
Embedding: A way of turning text into numbers so software can compare meaning.
Vector database: A database designed to search by meaning, not just exact words.
Retrieval: Finding relevant information before answering or doing a task.
RAG: Short for retrieval-augmented generation. The AI looks up relevant information first, then uses it to produce a better answer.

OOMU Control: The paid, organization-level part of OOMU: a server your organization runs itself. It sets which models every Mac may use, shows what each Mac is actually running, collects security events for your security team, and enrolls Macs through the device management you already use. The OOMU app itself is free to use, including at work.
Air-Gap Mode: A switch in OOMU that stops every outside connection. Cloud models, web search, and connected services pause, and all work stays on your Mac. An organization using OOMU Control can turn it on for every Mac at once.
Visual workflow: A workflow you can build and understand through visible steps.
Local-first work: Routine AI work that happens on your own computer whenever possible.
Approval step: A moment where OOMU stops and asks before continuing, such as before it runs a command on your Mac.
Clear record: A plain history of what OOMU did during a workflow.
Mod: A self-contained capability package (a single .oomu file) you add to OOMU to give your assistant something it did not have: a house style, a domain guide, or tagged tasks that read the Mod's reference material and keep notes for a project. Mods carry content and short instructions, never code, and they run inside OOMU's safety model, not loose on your machine.
Connector: A way for OOMU to work with another app, file source, or service.

Closing

It's to give people enough language to understand what the product is doing, ask better questions, and use AI with more confidence.

That's part of trust, too.