Instinct GOAP Framework

You describe facts, actions and goals. Instinct GOAP decides what the agent does

next — and tells you exactly why it decided that.

Behaviour trees and state machines make you author the transitions. GOAP makes you

author the world instead: what is true, what each action needs and changes, what the

agent wants. The planner finds the path. Add an action and every goal that can reach

it starts using it, with no graph to rewire.

Two APIs over the same planner. The Flow API is one domain file plus action classes,

for new agents and gameplay code you own. The Command API emits a command enum you

translate yourself, for slotting the planner into an existing state machine or

command bus without rewriting it. Pick one per agent.

Genre-agnostic, because it plans over your data, not over a scene format. The four

sample scenes are deliberately different shapes: a chef running a kitchen queue, a

farmer with an energy economy, a guard on patrol with alert states, and a stalker

A-Life agent with survival drives.

Fully customizable in the sense that matters for a planner: facts, actions, goals,

costs and goal-selection policy are all your own C# — there is no visual graph to

fight and no asset format to conform to. Action behaviour is either an async method

or composed steps.

Honest about scale: every replan clones world state, roughly 34 KB per replan and

750 B per agent per tick on a 57-fact domain. Comfortable for dozens of agents.

Not built for hundreds on mobile, and the Performance page in the docs says so with

numbers rather than promises.

Documentation is a nine-page handbook, also shipped as one offline HTML page, plus a

Ukrainian guide. Every sample has tests you can read as executable documentation.