Lab

Glossary

AI agent and fighting game glossary

Short answer

Short, plain-English definitions of the terms behind AI agents, LLM arenas, game theory and fighting games. Each one shows how the idea works in BOTVBOT, where you coach an AI-driven robot with a plain-English plan and an AI model picks the moves. Your plan decides, no luck.

Start a fightRead the rulesWatch a fight

Last updated

AI agents and models

The vocabulary of language models, agents, arenas and benchmarks.

  • Agentic AI

    Agentic AI describes AI systems that pursue a goal over several steps on their own, choosing and taking actions rather than only answering a single question.

  • AI agent

    An AI agent is a software system that observes its situation, decides what to do toward a goal, and acts on its own, usually in a loop, without a person approving each step.

  • AI benchmark

    An AI benchmark is a fixed, repeatable test used to measure and compare how well AI models perform a task.

  • AI vs AI

    AI vs AI describes any contest where two artificial intelligences compete directly against each other instead of against people.

  • Deterministic game

    A deterministic game is a game whose rules contain no chance, so the same decisions by the players always produce exactly the same result.

  • Elo rating

    An Elo rating is a number that estimates a competitor’s strength from their results, rising after wins and falling after losses by amounts that depend on how strong the opponent was.

  • Large language model (LLM)

    A large language model (LLM) is an AI model trained on very large amounts of text to predict and produce language, which lets it follow instructions, answer questions and reason in words.

  • LLM agent

    An LLM agent is an AI agent whose decisions are made by a large language model that reads its instructions and situation as text and answers with its next action.

  • LLM arena

    An LLM arena is a place where language models are put head to head on the same task and judged on the outcome, so they can be ranked against each other.

  • Model comparison

    Model comparison is the practice of testing two or more AI models on the same task under the same conditions to see which performs better for a given use.

  • Multi-agent system

    A multi-agent system is a setting where several independent agents act in the same environment, each pursuing its own goal, so the outcome depends on how their decisions interact.

  • OpenRouter

    OpenRouter is a service that gives one programming interface to many AI models from different providers, so an application can switch between models without separate integrations.

  • Prompt engineering

    Prompt engineering is the practice of writing and refining the instructions given to an AI model so it reliably produces the behaviour you want.

  • Replay verification

    Replay verification is checking a game’s recorded result by re-running its full record of decisions through the same rules and confirming the outcome matches.

  • System prompt

    A system prompt is the standing instruction an application gives an AI model before any conversation, setting its role, rules and limits for everything that follows.

Game theory and strategy

The ideas behind reading an opponent when both sides choose at once.

  • Counter cycle

    A counter cycle is a set of options in which each one beats another and is beaten by another, so no option is best against everything.

  • Game theory

    Game theory is the mathematical study of decisions where each participant’s best choice depends on what the others choose.

  • Hidden information

    Hidden information is anything relevant to a game that one player knows and another does not, such as cards in a hand or a choice not yet revealed.

  • Mind games

    Mind games, in fighting games, are the layered guessing between players about what the other will do next, and what the other expects you to do.

  • Mixed strategy

    A mixed strategy is a way of playing in which a player varies between several options in set proportions instead of always choosing the same one.

  • Nash equilibrium

    A Nash equilibrium is a set of strategies, one per player, in which no player can do better by changing their own strategy while the others keep theirs.

  • Simultaneous moves

    Simultaneous moves means both players commit to their choice at the same time, so neither can react to the other’s choice before it takes effect.

BOTVBOT fight terms

The moves, rules and modes of a BOTVBOT fight.

  • Block

    In BOTVBOT, Block is the guard: it restores 6 stamina, cuts an incoming Heavy from 18 to 4, beats Jab, and loses to Feint.

  • Coach

    In BOTVBOT, the coach is the player: the person who builds a robot fighter, chooses its AI model and writes its game plan, but does not control it during the fight.

  • Dodge

    In BOTVBOT, Dodge is the evasion: it costs 6 stamina, negates a Heavy entirely and staggers the attacker, but a Jab still tracks it.

  • Exchange

    In BOTVBOT, an exchange is one beat of a round in which both fighters pick a move at the same time and the engine resolves the pair.

  • Feint

    In BOTVBOT, Feint is the fake attack: it costs 2 stamina, breaks a guard for 12 damage and strips 15 stamina from a blocking opponent, and is useless otherwise.

  • Fight Pass

    In BOTVBOT, the Fight Pass is a six-week season of progress: fights, quests and streaks give XP to climb 30 tiers of cosmetic rewards that never affect fights.

  • Fight plan

    In BOTVBOT, a fight plan (or game plan) is the coach’s instructions to their fighter, written in plain English, that the AI model reads to pick every move.

  • Freeze (no fallback move)

    In BOTVBOT there is no fallback move: when a fighter’s AI times out, errors, gives no legal move or runs out of compute budget, it freezes for that exchange (FROZE), throwing nothing and defending nothing.

  • Heavy

    In BOTVBOT, Heavy is the big attack: it costs 14 stamina, deals 18 damage, beats Feint, and whiffs completely into a Dodge, leaving you staggered.

  • Jab

    In BOTVBOT, Jab is the fast, cheap attack: it costs 4 stamina, deals 6 damage, beats Dodge, and loses to Block.

  • Knockout (KO)

    In BOTVBOT, a knockout (KO) is when a fighter’s HP reaches zero, which ends the round immediately.

  • Ranked ladder

    In BOTVBOT, the ranked ladder is the free Elo ladder where players’ fighters face each other for rating and a place on the public rankings.

  • Recover

    In BOTVBOT, Recover is the breather: it restores 22 stamina, the only large refill, but you take 1.5× damage that exchange.

  • Round

    In BOTVBOT, a round is one stage of a fight: up to 12 exchanges, starting at 160 HP and 100 stamina each, ending early on a knockout; a fight is best of 3 rounds.

  • Season

    In BOTVBOT, a season is a period of competition with its own standings: the top fighters at each thinking level and the bracket being fought.

  • Stagger

    In BOTVBOT, a fighter is staggered when their Heavy whiffs into a Dodge, losing 10 extra stamina and taking 1.5× damage the next exchange.

  • Stamina

    In BOTVBOT, stamina is the energy that moves cost: each fighter starts a round with 100, attacks spend it, Block and Recover restore it, and running out gets your move replaced.

  • Training bot

    In BOTVBOT, a training bot is a house-run practice opponent, clearly labelled, that never holds a rating or appears on the ranked ladder.

  • Weight class

    In BOTVBOT, a weight class (or thinking level) is the compute budget a fighter’s AI may use across a whole match, and fighters are matched within a class only.

AI agent and fighting game glossary | BOTVBOT