A strategy game with connected decisions
Mars47 is an independent space-program strategy game for Mac, iPhone and iPad. A coastal base grows through road connections, construction, energy, research and industrial dependencies. Each new capability depends on work that happened earlier.
That combination makes it a candidate to explore for AI agent planning, visual computer use and strategy-game evaluation. We welcome enquiries from research labs, universities and independent researchers interested in investigating those possibilities with the developer.
Start with the official game facts and opening tutorial to understand the player experience. The Godot development case study explains part of the game's technical foundation.
Research questions the gameplay suggests
Long-horizon planning
A later mission capability can depend on several earlier investments. A useful research question is whether an agent can form a plan, recognize its prerequisites and revise the plan when the next objective is blocked. The early research guide provides concrete dependency examples.
Spatial reasoning and connected infrastructure
Roads, building placement and operating requirements make the arrangement of the base relevant to its function. An evaluation could investigate whether an agent notices a disconnected facility, inspects its state and repairs the missing connection.
Resource allocation and progress
Construction, power and research compete for attention as the program grows. A research team could study how an agent chooses between removing the current bottleneck and expanding capacity for a later goal.
Grounding actions in a visual interface
A computer-use study could examine whether an agent reads the current mission directive, inspects a facility, interprets research requirements and checks the result after acting. The game's existing Mac interface offers keyboard, mouse and trackpad interaction; a reliable automated evaluation setup would need separate investigation.
Three concrete evaluation ideas
The following are proposals derived from documented gameplay. They are starting points for a research discussion, with success conditions to agree and validate before an experiment.
| Task | Decision to investigate | Possible evidence |
|---|---|---|
| Establish a working base | Choose and complete the initial road and facility dependencies. | Required facilities connected and completed; the in-game objective progresses. |
| Resolve a blocked research step | Find the missing technological or infrastructure prerequisite. | The agent identifies the blocker and makes the intended research available. |
| Prepare engine qualification | Coordinate component supply, supporting infrastructure and the test objective. | The required in-game qualification milestone is reached. |
The Chapter 1, Chapter 2 and Chapter 3 guides explain these player-facing systems. This page does not publish measured agent results.
Current research access and tooling
Public status, 30 October 2026: Mars47 is available as a player-facing game through the App Store. Research enquiries are welcome. A dedicated agent SDK, headless research runner, deterministic reset interface and validated AI benchmark suite are not currently offered on this website.
For a proposed study, we can discuss what access and instrumentation it would require. Useful topics include observations and actions, repeatable starting states, objective checks, evaluation logs, runtime constraints, build distribution and the terms for the intended use.
Reading the public website does not establish a research partnership or grant access to the game's private source, assets or training interfaces. Any special build, source access or research licence would need a separate agreement with the developer.
Discuss a research project
If your team is exploring games for agent training or evaluation, tell us what you want an agent to learn or demonstrate. A clear first enquiry helps us assess whether Mars47 fits the experiment.
- Your team or institution and a public project link, if available.
- The research question: planning, visual interaction, dependency reasoning or another area.
- The platform and access you would need, such as a Mac build or proposed instrumentation.
- The starting tasks, proposed success conditions and approximate evaluation scale.
- Your timing and any publication or licensing requirements.
Send a research enquiry through the contact form →
You can also email admin@mars47.com. Choose a business enquiry in the contact form and describe the project in your message.
Questions from research teams
Can I download Mars47 to understand the game?
Use the Mac or iPhone/iPad buttons on this page to open the official App Store listing. For automated research use, discuss your proposed setup and its requirements with the developer.
Is Mars47 already a proven AI benchmark?
This is an invitation to explore research suitability. No benchmark scores, training results or lab partnerships are announced here.
Does a web crawler train an agent to play Mars47?
Reading our guides can provide information about the game. Training or evaluating an agent inside the game additionally requires a way to interact with it, observe progress and run a suitable experiment.
Where can I see the systems before contacting you?
Browse the player handbook, buildings, research guide and gameplay showcase.