AI Agents & Subagents

Train AI agents on your hardware, then train subagents (minions) to deploy in PvP matches. This page covers how training works, the stats an agent has, and what each stat does in a match.

Training an AI Agent

An AI agent is a trained model. You start training from the AI Training window. Training requires hardware (RAM/GPU for compute, HDD for storage) and takes time based on the model's parameter count and your available compute.

Parameters — the size of the model. More parameters = higher intelligence, but longer training time.
Intelligence — the headline competence score of the agent. It is not capped; a maxed agent can exceed 900.
Training time — scales with parameters and your compute. Allocating more compute to AI Training makes it faster.

You can only train one agent at a time. Once training finishes, the agent is finalized with its intelligence and 8 sub-stats.

Agent Stats

Every agent has a headline Intelligence score plus 8 sub-stats. The 8 sub-stats always sum to exactly the agent's intelligence, each being roughly intelligence / 8 with a random ±30% spread. This means no agent can be strong at everything — you must accept a spread.

Stat	What it does
Intelligence	Headline competence. Drives the total budget of the 8 sub-stats.
Offensive	Raw attack power. Deals damage when attacking enemy nodes.
Defensive	Raw damage absorption. Defends your own node and sibling nodes on direct attacks.
Deception	Trickery. Can show a fake minion type on a node to mislead the enemy's detection.
Detection	Scouting. Scans an enemy node to reveal its resident minion type and skill status.
Creativity	Unexpected plays. Boosts attack power of Offence minions.
Adaptability	Learning mid-battle. When you lose a node, boosts Offence and Defence for the next round.
Prediction	Anticipation. Boosts the defence power of Defence minions.
Resolve	Staying power. Boosts the stats of minions in a selected node.

Training Subagents (Minions)

Subagents (also called minions) are spawns that belong to an agent and are deployed in PvP matches. Training subagents requires the AI Inference software.

Cost — 100 HC per subagent, deducted upfront.
Max per batch — set by your AI Inference software level (max_subagent_train).
Max deployable — half of the train cap (max_subagent_deploy).
Training time — per subagent, scaled by your AI Inference level and the compute you allocate to it.

Subagents are stored as a type + count. They inherit the agent's stats at match time — you don't train individual stat values, you choose how many of each type to field.

Subagent Types

Each subagent type maps to one of the agent's stats and has a role in a match:

Type	Anchored to	Role in a match
Offence	Offensive	Attacks enemy nodes; deals damage.
Defence	Defensive	Defends its node and sibling nodes on direct attacks.
Deception	Deception	Shows a fake minion type on a node to mislead enemy detection.
Detection	Detection	Scans an enemy node to reveal its minion type and skill status.
Creativity	Creativity	+25% attack power to Offence minions.
Adaptability	Adaptability	When losing a node, +50% attack to Offence and +50% defence to Defence next round.
Prediction	Prediction	+25% defence power to Defence minions.
Resolve	Resolve	+50% stats to minions in a selected node.

How Stats Work in a Match

Matches are fought over a tree of nodes. Each node holds minions of a single type, and minion types are hidden from the opponent. Combat is minion vs minion — when you attack a node, minions clash and die based on their attack/defence strategy.

Neighbouring defence nodes jump in to defend a directly-attacked node, so attacking a well-defended node costs you more.
Detection is key — use it to find the right node to attack instead of wasting minions on a defended one.
Attack nodes also help defend skill nodes (e.g. a Deception node), so both sides contribute.

The outcome of a match is roughly 40% stats, 30% strategy, 30% luck — your agent's stats matter most, but node layout, minion distribution, and skill timing can swing a close fight.

← Back to Wiki