Run humanity's last shelter.The people inside remember.

A single-player, first-person survival and management game in development. You bring a shelter back to life after seventy years: restore its fixed machines, manage its resources and bring Deepborn to life, characters whose dialogue is generated during the conversation, with their personality, emotions and memories in context.

The project combines first-person survival and management with characters who can respond in generated dialogue. The prototype already includes bunker machines, resource information and an incubator, alongside conversation systems. The funded next step is to connect and develop those foundations into the Vertical Slice, not to claim that the full game is already complete.

Yes. THE BUNKERS 2108 is a single-player game.

Windows, on Steam first. The rest is not decided, and I won't announce it before it is.

The game's main language is French, with English subtitles planned. Additional languages are part of the Beyond stretch goal. The film at the top of this page is in English (Meet the Deepborn dubbed in post-production), and Meet the Deepborn is also offered on its own in French, with the voices recorded in the engine; the films do not change the game's languages.

No demo and no playtest are announced.

Keyboard and mouse are targeted at launch. Controller support is being studied and is not promised. Steam Deck is not an announced target.

Early Access is targeted 12 to 18 months after the campaign, and version 1.0 of Act 1 18 to 24 months after the campaign. With margin, deliberately. No firm release date and no Early Access price have been announced.

Yes. You can speak with a microphone or type on the keyboard.

No. You reactivate fixed machines, you don't build freely. The game is the people inside.

Sixteen years on it. I looked elsewhere, I came back.

Not in the setup demonstrated here. The current prototype uses online services for dialogue and voice. Local-first operation is a development objective, with a hybrid approach if necessary. Fully offline play is not confirmed.

Minimum and recommended specifications have not been announced, and compatibility with an RTX 3060 is not confirmed. A local language model running on a GPU by itself would not prove that the whole game runs well on that PC. Graphics, simulation, dialogue and voice need to be evaluated together. The captures on this page show the current prototype, not a local-performance benchmark.

That is not confirmed. The current prototype uses an online dialogue and voice pipeline. Moving some or all of that work onto the player's PC may involve trade-offs in response time, quality and hardware requirements. The final balance is still being developed.

The final requirements for online services and any related player-facing costs have not been announced. The current prototype's online services have usage costs; that does not by itself define how the released game will work. No subscription, personal API key requirement or unlimited hosted service is promised on this page.

The current dialogue and voice pipeline depends on online services. Its final fallback behaviour and offline availability have not been confirmed. Reducing that dependence is one of the reasons for the local-first development goal.

The current prototype relies on online services for parts of the dialogue and voice pipeline. The released game's data handling and retention arrangements are not finalised. This website's Privacy page describes this website, not a promise that all game conversations stay on your device.

No. Eight simultaneous Deepborn is a prototype test reported as of October 2026, not a promise of eight concurrent local conversations or a performance guarantee for a particular PC. The media on this page show the specific interactions described in their captions. The final population limit is not announced.

In the game, the Deepborn's dialogue and voices are generated during the conversation. For the campaign, the film Enter the world and the concept art are AI-generated or AI-assisted. Running the AI on your machine is an objective, not a guaranteed feature. The full disclosure is in the Use of AI section.

Every image on this page carries a label. CONCEPT ART illustrates the intended visual direction; it is not a capture of the game. AI-GENERATED STYLISED FILM marks Enter the world, a film that tells the world. ANIMATED EXPLAINER marks the Vision Deepborn diagrams. The film at the top, Inside THE BUNKERS 2108, carries one label for its three parts, and each chapter states what it is. DIRECT ENGINE CAPTURE · DEVELOPMENT BUILD is reserved for images recorded directly in the Unreal Engine development build (captures of October 2026): the environment and the French interface are work in progress, and nothing inside them is generated, removed or retouched. Their English translations sit under each image.

Yes. Deepborn dialogue is generated during the conversation rather than picked from pre-written lines. In Nola's captures on this page, the answers were not written in advance. Meet the Deepborn, at the start of the film at the top of the page, follows the same rule: the interview questions are prepared, the answers are not; its English version is a dub added in post-production. It is an edited film, so it does not show how long each answer took.

Deepborn use stored character information and memories as context for later dialogue. The prototype also has a catalogue of in-game books that can add authored knowledge. This is not a claim that a new foundation model is trained during each conversation, or that characters remember every word forever. The recall shown on this page does not, on its own, demonstrate memory after saving and restarting the game.

Generated dialogue can be mistaken or inconsistent. The prototype already supplies character information, memories and shelter context to conversations. More advanced handling of contradictions and emerging social situations remains part of development. It is not presented here as a finished, error-proof system.

The design centres on situations and needs within the shelter, rather than an unlimited quest generator. A character talking about an action is not proof that the action has been performed. The captures show interactions available in the current prototype; the broader connection between dialogue, situations and actions is still being developed.

No. Deepborn are simulated game characters. Personality, emotion and memory describe gameplay systems, not a claim of consciousness.

The films don't have the same status, and each one carries its label. Enter the world (first called Trailer 1) is a stylised, AI-generated film that tells the world. Vision Deepborn is an animated explainer, not engine footage. Meet the Deepborn (first called Trailer 2), at the start of the English film and on its own in French, was filmed on the stage built inside the prototype and then edited: the Deepborn's exchanges come from the AI dialogue and voice pipeline that is the principle of the game, today through online services. The French version keeps the voices recorded in the engine; the English version, in the complete film, uses an English dub added in post-production, so its voices are not the recorded ones. It is not a latency benchmark, a proof of memory after a restart or a demonstration of the complete game loop; how it was recorded is described in the AI disclosure under Use of AI. The vision behind the game, Nano's presentation between those two films, uses concept visualisations of systems and intentions.

Damiano Cibin, known as Nano, in Switzerland. Sixteen years on Unreal Engine, alone on the engine. Two friends have helped occasionally with some assets.

On this site, and on the project's Kickstarter profile, YouTube channel, Instagram account, Steam Community profile and Discord server.

Not yet. The Kickstarter campaign is not open, and the rewards on this page are planned. For now you can follow the project, for example on its Kickstarter creator profile. When the campaign opens, you will choose your reward and pledge on Kickstarter itself, never on this site.

Yes, on Kickstarter itself: you choose your reward and pledge there, never on this site. Kickstarter is all or nothing.

No. The Kickstarter campaign has ended, and no pledge is open. Its page stays on Kickstarter.

Because it is the minimum for six months of full-time development, minimal audio, the English translation, licences and campaign costs. The breakdown is in the "Why 45,000 CHF?" section above.

The 45,000 CHF gross goal funds six months of full-time development toward the Vertical Slice. It does not, on its own, pay for the finished game. Local-first AI is a development objective within the project's technical roadmap; it is not a guaranteed result demonstrated by the current media. The scope and target windows are explained in the funding and roadmap sections.

Kickstarter is all or nothing: you are not charged. If the goal isn't reached, Kickstarter doesn't fund the Vertical Slice. The project continues on my own funds, at my own pace, as it has from the start (eight months as of October 2026).