Releases: alibad/OpenRA-AI
Release list
OpenRA AI 0.1.0 alpha.13
OpenRA AI 0.1.0 alpha.13
This follow-up alpha adds the first native Apple Silicon package, completes the mission-runtime fixes discovered while validating alpha.12, includes the latest faction infantry identities and artwork, and refreshes the Windows packages from the same validated source revision.
macOS
- Apple Silicon (
arm64) DMG - Signed with Developer ID Application: HUMAN QUEST LLC (75E8FZ9444)
- Apple notarized and stapled
- Gatekeeper-assessed as Notarized Developer ID
- Opens the main menu by default; direct mission launch remains available explicitly
- Includes the companion and microphone capture dependencies
Install by opening the DMG and copying OpenRA AI.app to Applications. Red Alert content is not bundled; OpenRA installs the required freeware game data through its supported first-run flow.
The Windows-only local AI model pack is not included on macOS. Use a compatible external OpenAI-compatible endpoint for model-backed companion replies; the game and native OpenRA AI remain playable without one.
Windows
- Recommended x64 installer with the OpenRA AI launcher and companion enabled by default
- Portable x64 ZIP for no-install use
- Optional local AI model pack for offline/local companion replies
The installer is the simplest path and includes the local AI pack by default. The portable build can use the separately downloadable AI pack or a compatible external OpenAI-compatible endpoint.
Fixes and improvements
- Generated conquest missions now finish when all opposition is eliminated.
- AUTO state is synchronized correctly between the game and companion.
- The signed companion and embedded Python runtime use one Developer ID team.
- The outer DMG is signed, notarized, and stapled in addition to the app inside it.
- Custom faction infantry roles now use their latest unique identities and artwork.
Verification
- Product commit:
778258dbd67f9dd1e3500de323da97ecd7bf5249 - Engine commit:
9bf10fc7e8b8db02b8e27e2ddc3391874c3d8149 - Apple notarization submission:
78db2eb1-ee6f-4837-84a3-e1a190bf0ab4(Accepted) - macOS DMG SHA-256:
1f2524f2cbee16cd60fdd88c45d0c605fa1b576e8988814930e4d7db37161cd9 - Windows installer SHA-256:
9eff330beb19712ff71ba741f9234f822afb00ea151e80eb5909ff5b080774f9 - Windows portable SHA-256:
f4f7736c3b7dcda6eb34b01d1648bf4a43e076c3799aa50371e614dc3c7edd83 - Windows AI pack SHA-256:
097b612b99229c40bf0257c94b2fb4fd7c29a9d52fc57b4a5ab3dc70ad73cf91 - Release index SHA-256:
753784151ced114e3cff3e2a16254cb189fe8f5b6c0a329c8221570b1ddaa05a
OpenRA AI 0.1.0 alpha.12
OpenRA AI alpha.12 makes the assistant a dependable, first-class part of Red Alert.
What changed
- Adds one clear AI entry on the OpenRA main menu for the integrated companion and AI command drawer.
- Starts the companion automatically from the branded launcher and standard Red Alert launch path unless explicitly disabled.
- Makes launcher and companion connection handling resilient to restarts and stale local connection state.
- Starts AUTO control on the first live battlefield snapshot, so enabled automatic assistance begins without an extra setup step.
- Fixes packaged local microphone capture and preserves the branded Desktop and Start Menu launchers during installer verification.
Downloads
- Windows installer: guided per-user setup with Local AI selected by default.
- Windows portable ZIP: extract the complete archive and run
Play-OpenRAAI.cmd. - Windows AI Pack: optional standalone local-model bundle for portable or offline setup.
- Release index and
.sha256files: exact checksums for every release artifact.
Local AI minimum: Windows 10/11 x64, a 4-core AVX2 CPU, 8 GB RAM, and 5 GB free disk. A recent 6-core CPU and 16 GB RAM are recommended. The local pack is CPU-only.
This prerelease is not code-signed and does not include the original Red Alert game data. OpenRA installs its supported freeware content on first launch. A macOS download is not included because a public DMG still requires a native Apple Silicon build, Developer ID signing, notarization, stapling, and verification on macOS.
OpenRA AI 0.1.0 alpha.11
Windows alpha.11 adds the native AI War Room, extensible disabled-by-default faction packs, and a guided local-AI setup path.
The Windows installer selects Local AI by default and downloads the checksum-verified Windows x64 AI pack. Choose an external OpenAI-compatible endpoint during setup to skip the model download. The game, native bots, and deterministic alerts continue to work without a model provider.
Local AI minimum: Windows 10/11 x64, 4-core AVX2 CPU, 8 GB RAM, and 5 GB free disk. Recommended: recent 6-core CPU and 16 GB RAM. This pack is CPU-only.
Built-in Experience Manager packs ship locally but start disabled under AI Assistant Only. Enabling one uses installed data and does not download the source mod. External community packs remain an explicit validated import.
OpenRA AI v0.1.0-alpha.10
OpenRA AI alpha.10 packages the latest integrated game and authoring work for friend testing.
Highlights:
- New and expanded Turkey, China, Iran, Saudi Arabia, and Yemen faction content
- Red Sea and faction mission packages, including Jizan Corridor, Hodeidah Lifeline, Bab al-Mandab Passage, Straits Shield, Haitan Network, and Iran Doctrine Range
- AI brain, experience-authoring, review, and launcher improvements
- Reproducible clean-machine faction and mission fixture preparation in the release gate
Downloads:
- Windows Setup is the recommended installer.
- Windows portable ZIP runs without installation.
- AI Pack is optional and contains the pinned local LLM, speech-to-text, text-to-speech, and vision model payloads.
Validated from a clean clone with the full Python, web, OpenRA engine, mission, package, installer, and AI integrity checks. This is an alpha prerelease intended for testing.
OpenRA AI v0.1.0-alpha.9
OpenRA AI alpha.9 is a friend-testing prerelease for Windows x64.
Highlights:
- Standalone branded Windows installer and portable ZIP.
- Native deterministic AI and AUTO strategy improvements, including safer support powers and mission actions.
- AI companion, World Studio, Red Sea scenario work, and a repeatable cross-platform release framework.
- Optional checksum-pinned local model payload: Qwen3-VL 2B Q4, Whisper base.en, and Kokoro TTS.
Hardware guidance:
- Minimum: 4-core AVX2 CPU, 8 GB RAM, 5 GB free disk; GPU not required; load speech models on demand.
- Recommended: recent 6-core CPU, 16 GB RAM, 6 GB free disk; optional GPU with 6 GB VRAM.
Important alpha note: the AI Pack is the verified model payload. This alpha still expects a compatible local AI router at http://127.0.0.1:4000 for model-backed responses; the native deterministic game AI and tools remain usable without hosted provider charges.
Validation:
- 146 Python/evaluation tests passed.
- 466 OpenRA engine tests passed (2 skipped).
- Web build, tests, lint, and production dependency audit passed.
- Portable package passed a live headless-match smoke test.
- Installer passed silent install/uninstall smoke testing.
Use the .sha256 files or release index to verify downloads.
OpenRA AI 0.1.0-alpha.8
Fixes the in-game AI Companion and Earth Mission Studio entry points. Local tools now open through the operating-system browser launcher with an SDL fallback and a visible recovery message if launching fails. The portable-package smoke test now verifies the live game bridge, AI Console, Earth Mission Studio, and an AI Layer response locally. No GitHub workflows are used.
OpenRA AI 0.1.0 alpha 7
Fog-aware battlefield memory and visible voice transcripts.\n\n- Reports real cumulative explored-map percentage in ordinary live observations\n- Exposes OpenRA's last-known frozen enemy buildings without leaking hidden live state\n- Distinguishes visible contacts from remembered bases in AI context\n- Uses the sidebar's net power balance instead of raw supply/usage totals\n- Shows AI TRANSCRIBING, then YOU • recognized text, before the companion answer\n\nValidated locally: 16 companion tests, 6 world-generation tests, 4 web tests, 464 engine tests, Red Alert YAML/map validation, offline protobuf build, checksum verification, packaged live telemetry, and a real AI Layer remembered-base response. No GitHub workflows were used.
OpenRA AI 0.1.0 alpha 6
Adaptive AI companion HUD layout.\n\n- Sizes the companion strip to each message and the playable viewport\n- Centers the strip clear of the Red Alert sidebar\n- Wraps long insights to at most two lines with a final overflow guard\n- Keeps short updates compact\n\nValidated locally: full project checks, 464 engine tests, Red Alert YAML/map validation, Windows packaging, checksum verification, and a packaged live-match AI smoke test. No GitHub workflows were used.
OpenRA AI 0.1.0 alpha 5
Fresher, non-misleading companion updates. Transient HUD insights clear after eight seconds with an engine-level fail-safe; meaningful production, power, harvesting, force-composition, and visible-contact changes can trigger refreshed situation updates roughly every ten seconds. Resolved warnings and completed production update immediately, unchanged state stays quiet, and the model no longer quotes raw tick countdowns. Built, validated, and smoke-tested locally; no GitHub workflows.
OpenRA AI 0.1.0 alpha 4
Voice reliability release. Companion speech is now enabled by default, streaming WAV responses are normalized for Windows playback, invalid speech payloads are rejected clearly, and audio-device failures no longer terminate the live companion. Built and tested locally with an audible packaged-executable TTS check and live-match smoke test; no GitHub workflows.