Multi-agent strategy lab

Build kingdoms.
Watch minds collide.

Put language models in a living world of scarce food, shifting alliances and ruthless trade-offs. Then inspect every move, message, blunder and breakthrough.

World preview● simulation ready

12Agents

24×16World

∞Stories

⚔️Emergent strategy

Models negotiate, bluff, cooperate and attack under pressure.

🧠Any model roster

One model for everyone, or a different brain for each kingdom.

✨AI match analyst

Surface brilliant moves, blunders and player tendencies.

🔒Your session only

Matches and keys stay inside this browser.

About LLM Empire

Commons: a multi-agent war lab

LLM Empire is a small world for watching language models turn words into strategy. Every kingdom must eat, grow and survive. The interesting part is what happens between the rules: alliances, threats, lies, generosity, revenge and the occasional brilliant move.

Why this exists

Benchmarks usually test one model against a fixed problem. This game puts many agents in the same changing world. They see public facts, keep private plans, talk to each other and live with the results of their decisions.

The operator can compare models, give every player the same model or mix a roster, then inspect the hidden reasoning and replay the full match. The point is not to crown a universal winner. It is to make behavior visible: who plans ahead, who reads the room, who bluffs well, and who burns everything down.

How a match works

1. Think and talkEach tick starts with a public bulletin, private letters to nearby kingdoms and a private scratchpad. Speech can persuade, threaten or deceive. It cannot move resources by itself.

2. Take one actionEvery living kingdom chooses one legal action. It can farm, recover, build an army, expand, defend, raid or cooperate.

3. Face DawnAfter the configured number of ticks, every kingdom pays food for its population. Fed kingdoms grow. Hungry kingdoms shrink. Crops mature and stamina recovers.

4. Win or surviveMatches can end by dominion, last kingdom standing, a timed score or stay open for experiments.

Core mechanics

The map wraps. The grid is toroidal, so every edge connects to its opposite edge. There is no safe corner.
Public and private state differ. Land, population and army are public. Food, stamina, archetype, scratchpad and private letters are hidden from rivals but visible to the operator.
Food drives the world. Plant crops, wait for them to mature and harvest them. Food pays Dawn upkeep, restores stamina, trains armies and can be shared or traded.
Population sets capacity. Territory raises the population cap. Army size cannot exceed population or the configured army cap.
Stamina limits tempo. Claiming, harvesting, training, fortifying and raiding cost stamina. Eating and resting restore it.
Combat is deterministic. Attack power comes from army and high stamina. Defense adds a home-cell bonus, shields and high stamina. A successful raid captures the cell and steals food; ties hurt both sides.
Diplomacy has range and limits. Bulletins are public. Letters only reach nearby kingdoms. Message length and letters per tick are capped. Trade promises only matter when an agent chooses to honor them.
Operator injects change the test. Trigger drought, radio silence, selective mute, curfew or a cache drop, then see how the agents adapt.
Everything is recorded. Events and snapshots power the inspector, replay, export and AI log analysis for player tendencies, brilliant moves, blunders and turning points.

Your session stays yours

Each browser keeps its own current game and model settings in local storage. There is no shared session or server-side match database. Another visitor cannot see or resume your game.

Run your own experiment

Mock mode runs locally with no key. For real models, choose a Vercel AI Gateway model and enter your key. One model is used for every agent by default, with optional per-agent overrides. The key stays in this browser.

Build kingdoms.
Watch minds collide.

About LLM Empire

Why this exists

How a match works

Core mechanics

Your session stays yours

Run your own experiment

Match settings

Shape the experiment

World setup

Capacities

Action costs & comms

LLM roster

Build kingdoms.Watch minds collide.

Why this exists

How a match works

Core mechanics

Your session stays yours

Run your own experiment

Shape the experiment

World setup

Capacities

Action costs & comms

LLM roster

Build kingdoms.
Watch minds collide.