Notices the change
New armor, a power deficit, a lost harvester, or a critically damaged unit—not a running commentary.
Game event → relevance gateAI GAME COMPANIONA second mind.
A second mind.
Never a second driver.
It watches the same match you do, speaks only when useful, answers on demand, and gives command back instantly.
COMPANION / LIVE OBSERVATION ONLY
POWER130 / 90CONTACTS2 NEWSECTOREAST 48,20
THREAT / HIGH
“Heavy armor is entering from the east. Your northern route is still open.”
1.4s
HOW IT BEHAVESHelpful enough to speak.
Helpful enough to speak.
Safe enough to act.
The companion combines deterministic game state with fog-respecting tactical views. It can explain, propose a safe order for confirmation, or delegate real-time play to OpenRA's native bot stack when AUTO is explicitly enabled.
Respects fog of war
The model receives only the compact observation already visible to you. Hidden enemies stay hidden.
Snapshot → AI layer → one lineYou retain command
Confirm individual actions, switch strategies by voice, or turn AUTO on. Interrupt speech or disable delegation at any moment.
Ask · confirm · delegate · take backGO DEEPER
From a useful sentence to safe command.
The in-match surface stays quiet. The capability contract underneath it is explicit, inspectable, and interruptible.
In match
Browse the full Capability Atlas Push-to-talk + tactical feed
Ask naturally, see what was heard, and keep a compact history of advice and orders.
See proof + boundary In matchPriority-aware alerts
Information, warnings, and critical threats are treated differently.
See proof + boundary In matchSafe action proposals
The companion can propose a bounded order without silently taking control.
See proof + boundary In matchFog-respecting dual-view vision
The AI receives battlefield context without receiving information the player has not earned.
See proof + boundaryBUILT TO EVOLVEThe experience stays stable.
The experience stays stable.
The models can change.
The game never talks to a model provider directly. A private AI layer owns model credentials and named capabilities, while deterministic OpenRA logic keeps economy, production, combat, and AUTO play moving without waiting on an LLM.
Read the architectureOpenRA enginefog-respecting snapshot
OpenRA AIrelevance + safe actions
AI layernamed model routes
strategyprovider or local modeltranscriptionvoice input routespeechspoken response route