A hologram you can play. Characters who answer back.
A hand-built holographic arcade cabinet running an original time-travel adventure — where the characters listen to your actual voice, think, and answer in their own words.
Portals of Eden sends the player across human history — and every era is a place where something has gone wrong, guarded by someone who won't simply be beaten.
You cannot win this game by fighting. The Ghost Pirate cannot be defeated with a weapon, and he cannot be reasoned with, negotiated with or won over. He has to be outsmarted.
So how do you outsmart something that can see, hear, and even think about you?
Everyone else can be reached. Earn trust, offer protection, learn a name, find out what someone is actually afraid of — characters change their minds about you, and they remember that they did. Combat exists and is deliberately the least rewarding verb in the game. The conversation engine, the disposition model and the NPC memory all exist to make the social path the rich one.
He is the exception. That is what makes him the boss.
The characters are not on a monitor. They stand in the air inside the machine, reflected off a curved mirror, in real volumetric space.
A concave display mirror bends whatever you show it. So the game renders the world pre-bent the opposite way — a real-time geometric correction that cancels the optics before the light ever reaches the mirror, calibrated to this specific cabinet and saved to disk.
The controls are a genuine arcade stick and buttons, with one modification I made myself: the ball on top of the joystick spins. Push it to move, spin it to look around. The button lighting is driven by the game. This is a spiritual successor to the holographic cabinets of the early 1990s — built the way that generation of machines wanted to be built, with thirty years of technology they never had.
Speak into the cabinet and the character in front of you hears your real voice, thinks about it on the machine standing in the room, and answers out loud in a voice of their own. Nothing is pre-scripted. Nothing leaves the building.
A local language model drives every conversation. No cloud service, no API bill, no internet connection required. The arcade machine is self-contained — you could run it in a room with no network at all.
Characters carry their own histories, moods and opinions, and their memory of a conversation survives death, respawn and travel between worlds. Insult someone in one era and they will still be cold with you later.
Lip movement is driven by the live loudness of the audio actually being spoken, per character, so a figure only moves its mouth while it is genuinely the one talking.
Convincing a character of something is not a dialogue tree. What you actually say changes their disposition toward you, and a sufficiently persuaded character will do things for you that no menu option offers.
Not on request, and not for just anyone — a character has to trust you first, and some of them have very good reason to have stopped. Ask the right one at the right moment and you will hear something rare. Ask the wrong one, and you may find out why they went quiet.
A vision model gives characters visual context about the scene in front of them, so what they say is in response to what is actually happening around the player.
Both worlds are complete and playing in the cabinet today — world, characters, conversation, objectives and an ending.
Click any world to browse screenshots taken from the cabinet.
The pipeline that produced the finished worlds — concept art, sculpting, rigging, retargeting, into the engine — is proven and repeatable. These twenty eras have art direction and boss design. They are the queue.
Concept art Pre-production imagery. Not gameplay, and not final.
One universal controller drives every non-player character — perception, disposition, states, moods, weapons and per-character triggers — so a new character is configured, not programmed.
A dedicated runtime pre-warps the rendered image to cancel the cabinet's mirror geometry, with live calibration and per-cabinet settings saved to disk.
Local speech recognition, a local language model and local text-to-speech, chained together with streaming so characters begin answering before they have finished thinking.