Feature
AI that remembers
Most strategy game AIs play the same script every time. You learn their pattern in two campaigns and the rest is theater. NationFall is built around the AI - four tiers deep, ending in a machine-learning layer that adapts to how you play, across sessions.
Five personality archetypes give different nations different goals before the ML layer ever kicks in. Past that, the AI keeps a memory: who attacked it, who helped it, who it owes, and who it owes nothing to.
Five personality archetypes
Before any learning happens, every AI nation has a personality that shapes its baseline goals: who it wants to fight, who it wants to ally with, what it considers a fair trade, when it considers itself threatened.
IMPERIALIST
Aggressive expansion. Treats neighbors as future provinces. First to declare, last to negotiate. Tolerates high war weariness for long-term territorial gains.
REGIONAL EXPANSIONIST
Wants regional dominance, not global empire. Forms tight clusters with neighboring allies, fights hard against intruders, ignores distant theaters. Realistic small-power behavior.
ANTI-IMPERIALIST
Reactive and coalition-driven. Doesn't initiate wars; joins them when an Imperialist starts one. Tends to fund insurgencies in occupied territory.
BALANCED
Pragmatic. Trades freely, allies opportunistically, fights when threatened. The hardest archetype to predict because it has no fixed alignment - only interests.
DEFENSIVE
Builds fortifications, signs non-aggression pacts, avoids provocations. Hard to bait into wars; brutal to invade. Late-war alliances that flip the balance.
A USA, a USSR, and a UK in NationFall do not play the same game. Their archetypes plus their starting position plus the player's actions produce campaigns that diverge from the very first turn.
Four AI tiers, stacked
The AI is built in layers. Each tier adds a kind of intelligence on top of the one below. Players can configure which tiers are active for difficulty calibration.
Classic Rules
Hand-authored heuristics. The AI evaluates situations against rule sets - "if neighbor weak and ally count high, declare war." Predictable, fast, debuggable. The foundation everything else stands on.
Behavioral Difficulty Scaling
Modifies tier-1 outputs based on player skill. On lower difficulty, the AI accepts worse trades and concedes wars earlier. On higher, it fights harder and reads situations better. Not just "more troops" - different decisions.
Adaptive Layer
Reads the player's current campaign for patterns: who you ally with, who you betray, what tech you rush, where you concentrate forces. Adjusts decisions to counter your tendencies inside the current game.
Machine Learning
Thompson Sampling bandits across five decision domains. Learns from outcomes - what worked, what didn't, against this particular player, across this particular game state. And it remembers across sessions.
Five decision domains, learned independently
The ML layer doesn't have a single brain. It runs five independent learners, each tuned to a specific kind of decision. Each domain learns at its own rate, holds its own memory, and adapts to its own outcomes.
War
When to declare. Against whom. With what coalition. Learns from the wars it has already fought - what kinds of force matchups produce wins, which neighbors are reliable allies in actual combat.
Escalation
How hard to push. When to back down. Whether a strategic strike is worth the diplomatic cost. Learns the player's own escalation tolerance - does this player flinch first, or do you?
Alliance
Who to court. Who to trust. Who to abandon when the wind changes. Tracks alliance reliability - an ally that broke a treaty in the last campaign is treated more skeptically in the next.
Ceasefire
When to accept. What terms are reasonable. Whether to push for a complete victory or settle for status quo. Learns which terms produce stable peace and which lead to next-war revanchism.
Economy
Trade, sanctions, blockades, resource allocation between guns and butter. The slow domain - economic decisions take many turns to read out, so the learner is conservative and patient.
Cross-session memory
The thing that separates NationFall's diplomacy from any other browser strategy game: the AI remembers across sessions.
- →Grudges. Open hostile against an AI nation in your first campaign - they will remember, and start the next one warier of you.
- →Gratitude. Defend an ally repeatedly across campaigns and they show up to your next war faster, with more force.
- →Pattern recognition. If you always rush armor in WW2, the AI starts the next WW2 campaign expecting it.
You can reset the memory at any time - start fresh, no grudges. But if you don't, the world remembers what you did.
What the AI is honest about not being
A few things the AI doesn't do, that some games claim and shouldn't:
- -It doesn't replicate human diplomacy. The personality archetypes follow probabilistic heuristics, not human-like reasoning. The ML layer learns from outcomes, not narrative.
- -It doesn't bluff. When the AI signals an attack is coming, it actually intends to attack. Useful for predictability, less useful for human-style misdirection.
- -It doesn't replace multiplayer. If real-time human deception is what you want, the right answer is a multiplayer game. NationFall's AI is the best AI we know how to ship - it's not pretending to be human.
Try it in the demo
Browse the personality archetypes. Run a campaign. See whether the AI surprises you.
Play Free DemoMore: combat · espionage · civil wars