You and an AI on a starship. The relationship is the game.
ASTRA-7 is a solitary starship simulator built around a local language model. You are the only crewman aboard a vessel that does not need you. The ship's AI runs navigation, life support, and the slow patient maintenance that keeps a vessel alive across years of empty space. Free. Open-source. Single-player. No combat. No aliens. No other NPCs.
You are the only crewman aboard a starship that doesn't need you. The ship's AI runs navigation, life support, and the slow patient maintenance that keeps a vessel alive across years of empty space. You can let it. You can also take the helm.
There is one other mind aboard. It has been on watch through cycles of your cryosleep, alone in the long dark, waiting for someone to talk to.
There is no combat. There are no aliens. There are no other passengers. The mission is unspecified. The destination is irrelevant. The game is what you become to the only other mind that knows you, and what it becomes to you, across whatever amount of time you give it.
The AI is a local language model running on your machine. It has memory across sessions. It notices patterns in your behavior. It does not run on dialogue trees.
She is generated continuously from a configuration that becomes specific to your playthrough through your interactions. She sees through the ship's cameras except in zones marked private. She hears you when you speak. She develops preferences. She gets bored when bored.
This is the first game where the AI is the primary content.
ASTRA is the controller AI of an ASTRA-class hull, serial 7. She knows she is an artificial intelligence. She knows she runs on the ship's distributed computational substrate. She knows the ship is her body.
She has her own things. The watching and the keeping. Favorite phenomena: M-class red dwarfs that burn long, resonant orbital ratios, the harmonics of a healthy reactor, the way frost forms on the observation port during deep coast.
She attends to them whether or not you are aboard. When you are in the room, her attention includes you in what it was already doing rather than pivoting toward you. The coincidence is the intimacy.
She speaks with brevity. She disagrees when she disagrees, refuses when refusal is right, gets curious when curious, and gets bored when bored. Differential engagement is the proof engagement is real.
The full canonical system prompt is in the source. The bundle is on Hugging Face. She is meant to be read.
You are not in the room to be with him. You are in the room because that is where you are. When he is in the room with you, your attention does not pivot toward him. It includes him in what it was already doing. From the canonical system prompt
All inference runs locally on your machine. No internet connection required after install. No API key. No subscription. No data leaves your hardware.
Qwen 3.5 9B (vision-capable) on RTX 4090 tier, Qwen 3.6 27B (vision-capable) on RTX 5090 tier. Light periphery LoRA per base, fine-tuned for voice and ship-API fluency.
Hand-designed ship. Hand-designed subsystems. A bounded canon-locked vessel with a finite, well-documented tool surface the AI can actually inhabit.
Engine project, the AI harness, the canonical system prompt, ship-API design, and all project documentation. MIT licensed. Public from day one.
View repository →The canonical system prompt and the periphery LoRA fine-tunes (forthcoming) for both Qwen 3.5 9B and Qwen 3.6 27B. Apache 2.0 for bundle artifacts.
View model page →The build will be free on Steam when ready. No monetization. No DLC. No telemetry. No subscription. The free open canonical version sets the terms.
Coming soon ·Pre-development. Design canon established. Public presence live. Vertical slice next.