To the AGI

What is To the AGI?

To the AGI is a single-player sci-fi simulation and research tycoon set from June 2017. The starting position is deliberately small: one borrowed GPU, a garage workspace, and a plan to build a frontier research campus. The emphasis is on research and business decisions, with the lab as the central workspace.

The game mixes management and incremental progression with a fictional science campaign. Players choose what to train, weigh parameters and training tokens against hardware and cost, and turn useful models into income. Researchers generate insights, technology projects move faster, and future models improve. An isometric lab grows as compute expands, either through rented cloud GPUs or an owned fleet. The goal is to move from a tiny lab to a frontier research campus. The campaign frames these choices as a climb from a small lab to a frontier operation.

How to play

The opening experiment begins in Open Train: pick a model, set a token budget, and choose a compute option. Check the invoice before training, begin the run, and release the completed model. Different models can have different specialties, so the launch choice shapes what the lab can offer.

Income comes from API revenue. That money must cover salaries, data acquisition, and expansion decisions. Hiring researchers creates insights, which help complete technology projects faster and improve later models. The next objective points toward Campaign, where funding and science trials continue the story through nine trials and toward a fictional AGI synthesis.

Compute can be rented as cloud GPUs or built into an owned fleet. Balancing parameters, training tokens, hardware, and cost is the central planning task. The lab view grows as compute and facilities expand, giving a visual record of the operation. The next objective offers a practical guide when choosing among training options and budget levels.

Controls

Interaction is click or tap based. The lab can be dragged to change the view, and camera buttons provide another way to adjust the angle. Keys 1 through 5 switch views, while Escape returns to the lab. Fullscreen is recommended on desktop, where the isometric space benefits from more screen room.

The game runs in an HTML5 browser frame. Model training is simulated locally rather than through an external AI service, and a download, API key, or AI service account is not needed. That keeps the first session focused on the training screen rather than setup.

Browser storage holds progress, and wiping that storage erases the save. Hiding the tab pauses play, and offline income is absent, so training and revenue do not continue in the background.

Tips for your first run

The first training run is a budgeting exercise as much as a research choice. A cheaper compute option can keep the invoice manageable, but it may limit what the lab can attempt. Review the invoice before committing, and treat the finished model as a source of API revenue rather than the end of the experiment.

• Set a token budget after checking the invoice; parameters, tokens, hardware, and cost all interact.
• Use API revenue to hire researchers, acquire data, and expand the lab with a clear purpose.
• Follow the next objective and open Campaign when funding and science trials become the priority.
• Plan active sessions because the tab pauses when hidden and there is no offline income.
• Open Research, then Scaling & sources, for scaling-law equations, hardware references, and the interactive loss chart.
• Keep an eye on salaries and data costs as the lab grows, since API revenue has to cover both.

Content and replayability

The campaign is built around nine science trials and a fictional AGI synthesis. A separate achievement track offers 100 goals, which gives long-term management players concrete targets beyond the next model launch. Research projects and researcher insights feed back into future models, so the lab's progress compounds over time.

Model launches can specialize in different areas. The research area under Research, Scaling & sources includes published scaling-law equations, hardware reference material, and a loss chart that responds to input. These elements support planning rather than serving as pure decoration. Costs and progression include game assumptions, and the AGI ending is fictional.

The game is a playable prototype in active development. The opening, research balance, and interface are still being shaped, so the current build is an evolving management game rather than a finished campaign. Single-player structure and browser saving keep each run local.

Presentation and performance

The lab is presented as an isometric 3D space that grows with compute and facilities. The camera can be dragged or adjusted with buttons, and five view switches help inspect the operation. The browser frame runs the HTML5 build without a download, API key, or AI service account. The camera work supports inspection of the expanding lab.

Training is simulated locally in the browser rather than through an external AI service. Progress saves in browser storage, and clearing that storage removes the save. The game pauses when its tab is hidden, and offline income is not part of the design. Fullscreen is recommended on desktop for more room.

The research area adds a loss chart with interactive controls and hardware reference material, giving the numbers-heavy side of the lab a visual form. A playable HTML5 frame lets players inspect the camera and isometric space directly.

Who should play it

This is for players who enjoy management, incremental progression, and tycoon-style planning rather than direct action. The appeal is in reading a model-training budget, deciding between rented cloud GPUs and an owned fleet, and watching an isometric lab expand. Sci-fi framing and a campaign of science trials give the numbers a narrative direction.

Players who like achievement hunting have 100 goals to pursue. Those who prefer active sessions should note that the game pauses when hidden and offers no offline income. The prototype status also suits players interested in seeing an in-development research sim take shape.

The single-player focus and browser-based format suit players who want a local management project. It is free to play, so trying the opening experiment carries no purchase decision. The free browser build makes the opening experiment easy to sample before committing time to the nine trials and 100 achievements.