Skip to content

Glossary

Domain terms used across lAIfe, in one place. Each entry links to where the concept is explained or implemented. Terms are listed alphabetically.

Action

A single thing the brain decides to do: move, build, craft, interact, complete, or plan. An action is an intention, not a guarantee; a good one moves the mission forward, possibly across several turns. See entities/action.py and the game loop.

Brain

The LLM chain that, given an observation and the current mission, returns the next action. This is the core decision step of a player's turn. See the llm package.

Building

A placed structure in the world, such as a house, farm, or factory. A building is an instance; its category and fixed size come from its building type. See the entities package.

Building type

Master data for a category of building: its name, description, and size, shared by every instance of that category. See entities/building_types.py.

Judge

See world judge.

Mission

A goal a player works toward, with an objective, a status, and a history of actions and results. Missions can nest into sub-missions and are generated by an LLM rather than scripted. See llm/mission.py and the game loop.

Mission generator

The LLM chain that proposes a fresh mission objective from the current world observation, so a player without a goal can find one. See the llm package.

Observation

A structured spatial snapshot of the world centred on a player: the nearby entities with their type, name, relative position, and distance. It renders to text using cardinal directions so the LLM reasons in spatial language. See entities/world_map_observation.py.

Planner

The LLM chain that decomposes a hard mission into ordered sub-missions, using the mission history as context. Triggered by the plan action. See the llm package.

Player

An agent in the world, owning its position, state, inventory, and current mission. A player perceives the world, asks its brain for an action, and turns that action into a world request. See the entities package and the architecture.

Replier

The LLM chain that produces an in-character reply to a message from another player, without touching world I/O. See the llm package.

Sub-mission

A mission nested under a parent mission, produced by the planner when a goal is too complex to solve directly. See Mission.

Terrain

A region of the map with a type, such as forest, lake, or fertile land, used for prompt context and rendering. See the entities package.

Utensil

A tool a player can hold and use, such as a bucket, axe, hammer, or hoe. A mission may require a utensil that the player must find or craft. See entities/utensil.py.

World

The authority over all state: players, buildings, and terrain. A player changes the world only by sending a request and awaiting the answer; the world validates every request and never tells a player what to do. It is run by the world runner. See the architecture.

World channel

The typed contract a player and the world communicate over: requests and responses, with no loose dict payloads. See world request and world response.

World judge

The LLM chain the world uses to decide whether a build or craft action is valid, returning a success flag with feedback. This is the world's reasoning, distinct from the player's brain. See entities/world_judge.py.

World request

A typed message a player sends to the world to act: build, craft, move, observe, interact, or complete. Called WReq in the code. See entities/world_channel.py.

World response

A typed message the world returns for a request, carrying a status and a payload specific to the action, or an error. Called WRes in the code, with concrete subclasses per action. See entities/world_channel.py.

World runner

The object that holds world state and runs the simulation loop: it reads requests, validates them, runs collision and judging, mutates state on success, and returns responses. See the entities package.