Everything built.
At the depth you need.

The companion calls out meaningful changes such as a power crisis, new armor, a lost harvester, or an exposed attack window.

Deterministic game-state transitions pass through a relevance gate before any model is asked to phrase an update.

Hold Ctrl+Space on Windows or Option+Space on macOS to ask. The transcript appears in game, followed by a concise answer you can interrupt.

Audio routes through the configured transcription capability while the feed records advice, alerts, transcripts, orders, and execution results.

Low-priority context can appear quietly while profound or urgent changes may use voice and a stronger visual treatment.

A deterministic threat and novelty score assigns the notification tier before text or speech generation.

Review a proposed action, confirm with Ctrl+Enter, reject with Ctrl+Backspace, or simply ignore it.

Model output is translated into a small allowlist of validated OpenRA orders rather than arbitrary engine calls.

AUTO can handle economy, production, expansion, defense, and combat while the companion explains significant decisions.

The feature uses the full ModularBot implementation, with voice or hotkey strategy changes for aggressive, defensive, naval, or adaptive play.

Advice can account for the current viewport and remembered whole-map tactical state, including an attack already under way.

Compact structured state is paired with a viewport image and a fog-filtered tactical overview for capable vision routes.

Pin a landmark or region, inspect the source terrain, choose scale and mission shape, then create from the selected area.

The studio combines map interaction, geocoding, satellite-style imagery, coordinates, and a deterministic mission blueprint.

The generation pipeline shows what it is reading instead of hiding the source-to-map translation behind a spinner.

Geographic data and the exact terrain view can be routed through a vision model, then normalized into an OpenRA-safe terrain plan.

The result looks and behaves like an OpenRA battlefield rather than a picture stamped onto a grid.

A deterministic translator hands the terrain plan to OpenRA-compatible map generation primitives and packages map.bin, YAML, and metadata.

Validation reports route safety, connected starts, reachable resources, and legal map structure.

Tracked-unit movement rules are used to test paths between spawn regions and key gameplay targets after terrain generation.

Download the generated battlefield, install it locally, and reproduce the same terrain with the same mission blueprint and seed.

The browser-side compiler builds a zipped .oramap with map data, rules, actors, and a source manifest.

Start from a playable Earth-derived base, then refine terrain, actors, resources, and mission logic with native tools.

Generated packages preserve normal OpenRA map structure instead of introducing a closed proprietary format.

Play Jizan Corridor, Hodeidah Lifeline, and Bab al-Mandab Passage as Saudi Arabia or Yemen.

The theatre combines faction rules, objectives, bot priorities, localized voices, and source-dated mission contracts.

Coordinate Karrar armor, Raad air defense, Fajr standoff fire, reusable and expendable UAVs, coastal forces, and the fictional Shadow One commando.

Fourteen production roles use native OpenRA ammunition, rearming, stealth, targeting, naval movement, bot production, and prerequisite contracts.

Field sixteen original combat actors through a full technology tree and play the Straits Shield land-air-sea mission.

Original sprites, effects, voices, AI production rules, and one construction-enabled mission reuse OpenRA's native movement, rearm, naval, and bot systems.

Use network specialists to coordinate armor, air defense, amphibious forces, aviation, and naval units in Haitian Network.

Seventeen combat actors, original procedural effects, bilingual voice lines, bot rules, and native renderer evidence form a reproducible faction package.

Distinct rosters, directional vehicle sprites, faction-specific voices, UI art, effects, and battlefield evidence make every force feel authored.

Assets and rules live in the mod structure, are audited against OpenRA's directional, palette, and packaging requirements, and are indexed into the public build archive.

Future releases can introduce new factions, units, stories, voices, maps, and rules without rewriting the core platform.

Shared platform contracts separate companion capabilities, Earth generation, distribution, and theatre-specific content.

Developers can run repeatable experiments in which an agent plays a match, records evidence, and explains its decisions.

A local MCP server exposes a safe observation and action allowlist to an Agents SDK commander.

Each attempt leaves a readable trail: observations, actions, outcomes, grades, and lessons for the next run.

Run evidence is graded against explicit objectives and condensed lessons are injected into subsequent attempts.

Players can use hosted models for simplicity or local routes for privacy, latency, and cost control.

Companion text, terrain vision, transcription, and speech are independent routes behind one generic AI layer contract.

Download the pack separately, choose local capabilities, and keep core gameplay usable when no model is running.

The pack prepares routes for battlefield language/vision, transcription, and speech; each capability can be started independently.

Friendly presets keep the common path simple while advanced routing and masked credentials remain available when needed.

Settings map friendly provider/model choices to AI-layer routes and estimate text, transcription, and speech usage.

Download a versioned build, verify its checksum, install or extract it, and launch the companion or current theatre.

The web app discovers real release assets and exposes version, size, checksum, and local smoke-test status.

Sign in with Google or email, create a mission package, download the game, and report feedback with the relevant page context.

Firebase authentication gates AI-backed work while private analytics use pseudonymous account identifiers.

Download the disk image, copy OpenRA AI to Applications, and launch the same companion and creation experience available on Windows.

Release discovery exposes the published arm64 DMG, version, size, and checksum alongside the Windows packages.