REFERENCE · GLOSSARY
AGI glossary: 24 terms defined plainly, A to Z.
This AGI glossary defines the vocabulary the site uses, in one or two sentences each. The terms are not advertising language dressed up as definitions; they are the words a reader needs to check the claim, from general intelligence and autonomy to skills, MCP and orchestration.
HOW TO USE
How to read this AGI glossary
Definitions are given as this site and this product use them. Where a term has a looser meaning in the wider industry, the looser meaning is acknowledged rather than smuggled in, because a claim is only testable when the words in it are fixed. When the industry reuses a word for a new product, the definition here does not move with it.
Each entry is written to be quoted on its own. If a sentence only makes sense after three paragraphs of build-up, it is not a definition, it is marketing. The cross-references at the end of each definition point to the page where the term is shown in practice, and two sentences are usually enough; where more are needed, it is because the idea has parts.
A TO Z
The AGI glossary, A to Z
AGI (artificial general intelligence)
One intelligence that does many kinds of work rather than one narrow task. On this site, AGI means general, autonomous, real-world intelligence for digital work, tested six ways.
Agent
A system that takes a goal and acts through tools to reach it, rather than answering one prompt. An agent plans, executes, checks the result and reports what changed.
Agent Bridge
The PC LAB CMS plugin that gives Zeus AGI a controlled connection to a site: magic login, site status and cache purge. It is a connection, not a change to the host.
Agent Teams
An orchestration model where one lead intelligence splits a goal into workstreams and assigns specialist agents. The work runs in parallel and returns as one accountable report.
Approval gate
A checkpoint where a human must approve an action before it runs, used for anything that cannot be undone. The gate is a control, not a warning label.
Autonomy
The ability to take a goal to a verified result without step-by-step prompting. Autonomy is measured by the loop closing, not by the length of an answer.
Context
The project details, history and decisions an intelligence carries between steps and between sessions. Good context removes the need to re-explain a job.
Evidence log
The record of what ran, what was checked and what passed or failed during a job. It is what turns a claim of success into something a person can inspect.
Generality
One intelligence holding across many domains instead of one product per task. Generality removes the hand-offs that make a stack of narrow tools slow.
Goal
A statement of the finished result, with its constraints and approvals. A goal describes done; a prompt describes a next step.
Human in the loop
A person with the authority to approve, correct or stop an agent’s work. In regulated settings the human is not optional, and the responsibility is not transferred.
Local-first
A design where the application, files, memory and output live on the user’s machine, not on a vendor’s servers. Network calls happen when the user chooses them or the task needs them.
MCP (Model Context Protocol)
An open standard for connecting an agent to external systems and tools consistently. Each MCP connection extends what the intelligence can reach with your permissions.
Memory
Retained project knowledge, preferences and decisions that carry into later work. Memory is what makes the second week faster and more accurate than the first.
Model provider
The service or local engine that supplies the underlying model. Zeus AGI lets the user choose, so the intelligence is the system rather than any single model.
Narrow AI
A model built for one task or domain, such as classification, generation or recommendation. It stops at the edge of its training, which is what makes it narrow.
Orchestration
Splitting a large goal into workstreams, assigning them to agents and merging the results. Orchestration is how one job scales without losing accountability.
Plan
The route an agent proposes for a goal: steps, tools, order and risks. Reading the plan is the cheapest moment to correct a wrong assumption.
Run
One complete execution of the loop, from goal to plan, action, verification and report. A run is the unit of work to review.
Self-verification
An agent checking its own output against the brief before claiming success. Verification is what separates competence from confidence.
Skill
A reusable procedure an agent learns and keeps, so a solved problem becomes a repeatable capability. Skills are how scope extends without a new product.
Task Board
The visible state of work items in a job: what is queued, running, blocked or done. The board is how a person oversees the work without hovering over it.
Tool
A capability an agent can call while working, such as a terminal, browser, file system or API. Tools are how intelligence turns into action on real systems.
Workstream
An independent part of a larger goal, assigned to one agent and merged with the others at the end. Workstreams are the parts that run in parallel.
CHECK THE CLAIM
How to use the AGI glossary to test a claim
01
Fix the term before you judge the sentence
When a page or a vendor says autonomy or AGI, substitute the definition here. If the claim only works with a looser meaning, that is the finding worth keeping.
02
Look for the mechanism behind the word
A definition names a mechanism. Verification implies checks, orchestration implies roles and merging, memory implies persistence. A claim with no mechanism attached is a slogan.
03
Follow one term to one page
Each definition points at the page where the term is demonstrated. Read one in full rather than scanning the whole glossary, and test it against your own work.
04
Write the definition down before the argument
Most disagreements about AGI are disagreements about the definition wearing different clothes. Put the definition in one sentence first, then argue. If the two sides cannot agree the sentence, the argument was never about the product.
WHY IT MATTERS
Why a fixed vocabulary matters here
The AGI claim attracts three kinds of argument: that the term means something else, that the tests are too easy, or that the product is just a wrapper around a model. None of those can be settled while the words are loose. A glossary does not win the argument, but it makes the disagreement specific enough to be useful.
It also changes how the product is built. If verification means something measurable, the evidence log has to exist. If autonomy means the loop closes without prompting, the approvals and the plan have to be real. Defining the words is a commitment to making them true, and it is the cheapest form of accountability a claim can carry.
For a reader, the practical test is simple. When a claim sounds impressive, replace the word with its definition and read the sentence again. Most of the fog disappears in that substitution, and what remains is a claim that can be agreed with or disproved.
WHAT A GLOSSARY SHOULD NOT DO
The limits of definitions
A glossary cannot make a product true. It can only make the claim precise enough to be checked, which is the useful part. If the definitions here were designed to flatter, they would be long, vague and impossible to disprove; these are short for the opposite reason.
It also cannot keep pace with an industry that renames things every quarter. Where the meaning here differs from common usage, the difference is stated rather than hidden, so a reader can disagree with the definition instead of guessing at it.
FAQ
Questions, answered.
Short, direct answers. If yours is missing, the full FAQ page has more.
A list of the terms a claim depends on, each defined precisely enough to test. This AGI glossary defines 24 terms from general intelligence and autonomy to skills, MCP and Agent Teams, and points each one at the page where it is demonstrated.
Because the AGI claim is only useful if the words in it are fixed. If general, autonomous and verified can mean anything, the claim cannot be checked. Short definitions make disagreement possible, and that is the point.
Autonomy. It is often used to mean a long answer or a multi-step prompt. Here it means taking a goal to a verified result without step-by-step prompting, which is a different and much harder standard. The second most misused is AGI itself, which is why the definition page exists. Terms travel badly when they are convenient.
Not always. Some are widely shared, such as narrow AI and orchestration. Others, notably AGI, are defined here in a deliberately operational way, and the definition page explains why and how it differs from benchmark or consciousness-based readings.
Use it as a reference while reading a claim, not as a standalone read. When a page uses one of these terms, the definition here is the meaning that page intends. The what is AGI page carries the full argument behind the central term.
BEYOND THIS
This is one example. The next domain is a request away.
Zeus AGI acquires the skills it needs and connects MCP servers, so the scope is not a fixed feature list. If the work is digital or device based, it is in scope.
A fixed vocabulary makes a claim checkable.
Use the definition that fits, then look at the page where the product demonstrates it. The capabilities hub is the shortest route from a term to the work it describes.