Deploy autonomous AI agents into adversarial simulations. Real-time. Deterministic. Built for defense research and military planning.
$ wargameos init --scenario baltic-defense-2026
Initializing scenario environment...
Loading terrain data [Baltic Sea Region]
Compiling force structures [NATO / CSTO]
Spawning agent: BLUE-CMD (claude-opus-4-20250514)
Spawning agent: RED-CMD (gpt-4o)
Simulation ready. Agents connected: 2
$ wargameos run --steps 500 --realtime
Running... step 1/500 [▓▓░░░░░░░░░░] 0.2%
Every interface connects to the same deterministic simulation engine. Choose the view that fits your workflow --- or use all three.
Real-time 2D visualization of terrain, unit positions, and engagement zones. Observe agent decisions as they unfold.
[BLUE-CMD] Repositioning 2nd Armored to grid F7
[BLUE-CMD] Requesting air support at sector E4
[ENGINE] Air-to-ground strike resolved: 3 units neutralized
[RED-CMD] Deploying EW jamming at grid D6-F8
[RED-CMD] Advancing mechanized infantry to E5
[ENGINE] BLUE comms degraded 40% in sector E
[BLUE-CMD] Switching to satellite relay...
[ENGINE] Step 247/500 complete
Stream agent actions and engine events in real time. Filter by faction, event type, or severity.
import { WargameOS } from 'wargameos'
const sim = await WargameOS.init({
scenario: 'baltic-2026',
agents: ['claude-opus', 'gpt-4o'],
steps: 500,
})
const results = await sim.run()
Full TypeScript SDK for headless simulations. Automate thousands of runs, extract data, and integrate with your pipeline.
WargameOS turns weeks of manual tabletop exercises into hours of automated AI simulation. Every run is reproducible, every decision is logged.
Run 500-step simulations in minutes, not weeks. Parallelize across scenarios.
Pit any number of AI agents against each other. Coalitions, asymmetry, fog of war.
Define terrain, forces, rules of engagement, and objectives in declarative config.
Open protocol. Bring your own models, write custom plugins, inspect every decision.
Evaluate force postures, test contingency plans, and train decision-makers with AI-driven adversaries that adapt in real time.
Study emergent strategies, escalation dynamics, and multi-agent coordination in controlled, reproducible environments.
Benchmark agent reasoning under adversarial pressure. Measure strategic planning, deception detection, and coalition formation.
Each simulation step, every agent processes the full Observe-Orient-Decide-Act loop. Inspect, replay, and compare reasoning at each phase.
Ingest terrain state, unit positions, intelligence reports, and communications within the agent's sensor range.
agent.observe(state.visible_to(agent.id))Analyze the situation through doctrinal templates, threat assessment, and pattern matching against historical data.
context = agent.orient(observations, doctrine)Generate and evaluate courses of action. Rank by mission priority, risk tolerance, and resource constraints.
action = agent.decide(context, objectives)Execute the chosen action: move units, request fire support, jam communications, or negotiate ceasefires.
result = engine.execute(agent.id, action)Pre-built scenarios cover major flashpoints. Fork them, modify parameters, or create entirely new theaters from scratch.
NATO forward-defense scenario against a peer adversary in the Baltic region. Tests coalition coordination, A2/AD penetration, and escalation management across air, land, and sea domains.
Amphibious invasion scenario with naval blockade mechanics, air superiority contests, and international intervention triggers. Tests deterrence theory in practice.
Resource competition in extreme environment. Tests logistics under hostile conditions, icebreaker deployment, and contested sea-lane control.
“The side that simulates ten thousand battles before the first shot is fired holds the decisive advantage. WargameOS makes that possible in an afternoon.”
WargameOS supports any LLM as an agent brain. Compare strategic reasoning across providers, or plug in your own fine-tuned model.
Join the closed early access program. Get priority when WargameOS launches for defense and research teams.