[ Decision Guide ]

Is Intent Forge right for your project?

A frank answer in 60 seconds. Skim the green and red lights, then run the five-question check. Still not sure? The Quickstart takes 10 minutes and tells you more than any docs page can.

10-min decisionNo code

60-second answer

Intent Forge is a Goal-Oriented Action Planning framework for UE5. You declare goals (target world states) and actions (preconditions + effects); a planner invents the action sequence that gets the agent from where it is to where the goal wants it. The signature feature is the five-family anti-flap toolkit: agents don't visibly oscillate between near-equal goals the way most GOAP implementations do.

It's the right pick when you want planning with stability that plugs into whatever AI runtime you already use (or none at all).

Green lights

Red lights

Five-question check

1. Does your agent need to plan ahead — not just react?

GOAP shines when the agent has to sequence multiple actions to reach a goal whose preconditions aren't currently true. "To eat, I need food → to get food I need to cook → to cook I need ingredients → to get ingredients I need to forage."

Yes if your agent regularly needs ≥ 2-step plans. No if it picks one highest-utility action per tick — Utility AI alone is fine.

2. Does the goal mix shift at runtime?

GOAP wins when goal priority changes based on world state. A guard juggling Patrol, Investigate, Pursue, Eat, Sleep, and Report-to-superior, with the planner picking the right sub-plan, is the canonical shape.

Yes if your agent has ≥ 3 competing goals whose priorities depend on state. No if there's only ever one goal active and a state machine models it cleanly.

3. Do you author at the goals + atomic actions level?

GOAP is declarative. You write goal = HasFood = true and action = Forage { pre: NearTree, post: HasFood } and let the planner figure out the sequence.

Yes if you find yourself writing the same if condition then sequence boilerplate across many characters. No if designers want to script sequences explicitly — use a BT.

4. Do you want stability without writing your own anti-flap?

The five-family toolkit ships ready. If you've ever shipped an AAA-feeling project derailed by agents that flicker between near-equal goals — Patrol/Chase across a range boundary, Eat/Drink scoring 0.49 vs 0.51 — the toolkit is the pitch.

Yes if you want stable behaviour out of the box. No if you trust your existing stability solution, but read the toolkit anyway, you may find a layer you missed.

5. Do you want flexible execution, or none at all?

The dispatcher abstraction lets the same FPlan execute via the built-in subsystem (no AI runtime), via StateTree, via BT, or via your own external loop. Most GOAP plugins are coupled to one runtime; this one isn't.

Yes if you're building solo / indie / for an existing AI stack you don't want to fight. No if you're already locked into a specific runtime — Intent Forge still works; the dispatcher just lives where you tell it.

Alternatives

You don't have to pick one. Intent Forge slots into a BT or StateTree as a single task — use BT/StateTree for the outer transition logic (combat ↔ patrol ↔ death) and let Intent Forge plan the actions within combat mode.

Recurring criticisms

Anti-patterns

If you've decided to use Intent Forge, here's how not to use it.

Recommendation

If you're not sure, do the Quickstart and build the forager. 10 minutes will tell you whether the model fits your brain. The cost of trying is low; the cost of picking the wrong AI paradigm for a 6-month project is huge.

◢ Next

Try this next

Hands-on /Quickstart (10–15 min)

Empty project → running plan. Five data assets, one Blueprint, no C++.

Mental model /Concepts

Planner, dispatcher, executor, and why they stay decoupled.

Signature feature /Anti-Flap Toolkit

The five-family stability layer. The reason Intent Forge exists.