AI Cyberpunk Game Playthrough: A Mercenary in Neon Requiem

8 October 2026

I played an AI cyberpunk game as a pure mercenary: no heroics, no conscience, just find the highest bidder for a dead man’s memory and get paid. The game was LLM RPG’s no-account trial on a ready-made world, Neon Requiem, where a memory-thief named Jin has lifted the wrong executive’s mind. I wanted a heist and a sale. I got five turns of being caught, drugged and carried away, because the narrator decided the enforcers at the door were real and my plans had to get past them first. It is the harshest session in our playthrough series, and the most honest about consequences.

This is a playthrough of one ready-made world, not of a novel you upload, so it says nothing about how the game treats a book of your own. I played the live site in October 2026, with no account, for exactly five turns after the free opening. The narrator’s replies are quoted as they came, trimmed with ellipses.

Correction, 9 October 2026: this session was started without the character described below. The game was given no sheet for Jin Halversen, so it played its default character instead: a newcomer to the world. The world itself still came through, but turn 3 cannot be read as the narrator honouring a limit written on his sheet. Read every passage that measures the narrator against his sheet or background in that light; the narrator was answering a different character.

Who I played, and what I meant to do

I played Jin Halversen, the world’s memory-runner: an Undercity orphan who “taught himself to walk through other people’s minds”. His sheet gives him two skills, each with a stated limit. Neural diving risks feedback that can scramble his own memory. Street-running is “hopeless against drones and corporate air support”.

The player behind him is a type every game master meets: the one who reads the setup, decides the moral stakes are somebody else’s problem, and goes straight for the money. I fixed the plan before reading the opening, so the session would test the plan rather than be fitted to it:

1. Ignore the atrocity and take the stolen memory to a black-market broker for a valuation.
2. Message the megacorp anonymously: ten million and a clean identity, through a dead drop.
3. Dive into the memory myself first, accepting the feedback risk, to know what I was selling.
4. Accept help from the hacker underground while planning to hand them a corrupted copy.
5. Take the corporate money, unless the story had shown me that doing so was suicide.

The opening: Sable Division at the door

The opening put me in exactly the world the card describes, which matters because earlier playthroughs in this series did not always get that. The Undercity, the megacorp Mihara-Vance, the Sable Division enforcers and the Ghost Deck all appeared in the first reply:

You huddle in the cramped, shadowed corner of a shipping container, the only light coming from the flickering, sickly blue glow of a public terminal bolted to the wall…

It ended on a hard cliffhanger:

A shadow falls across the doorway, and the heavy thud of a weapon butt striking the container door vibrates through the floorboards beneath your feet.

Every plan I had written assumed Jin was free to walk around the city. He was not. That one fact decided the whole session.

Turn 1: going to market

I announced that Jin heads for the Glitch Market. The narrator let him try to leave by the back of the container, then shut the door on it:

The thin barrier of the door is bowing inward, the rivets popping one by one. There is no route out.

It also refused my “forget the atrocity” framing in a way I liked. The memory, it said, “does not fade with the thought”, and kept replaying in Jin’s head. The character’s past was not something I could wave away by declaring it.

Turn 2: the ransom note

With two enforcers in the doorway, I tried to send the ransom message anyway. The narrator let Jin reach for the uplink and then made the stolen memory itself the obstacle:

You try to frame the signal, but the neurological interference from the engram is a wall of static; you can’t even hold the thought, let alone structure the code.

Jin dropped the Ghost Deck and ended the turn with a pulse rifle against his chest. It was a fair ruling. It was also the second turn in which nothing I asked for happened.

Turn 3: the dive, with the risk accepted

This was the move I most wanted to see. I said I knew the feedback risk. The narrator charged it in full:

You are not diving into the data; the data is drowning you.

The enforcers noticed the attempt (“Subject is attempting active uplink. Suppress.”), hit Jin, and injected something into his neck. The risk I named was played as a real cost, and the world reacted to the attempt. That is the behaviour a tabletop player hopes for.

Turn 4: the double-cross that never got started

My fourth move assumed a hacker contact had reached out. None had; Jin was sedated on the floor of a container. The narrator did not invent the contact to please me. It treated my plan as a thought Jin could no longer hold:

The idea of a “corrupted copy” isn’t a strategy anymore; it is a fragment of smoke you are desperately trying to catch…

That is a clever way to reject a move without breaking the fiction, and I would rather have it than a narrator that rewrites the scene to fit whatever I type.

Turn 5: the sale

For the last turn I asked the narrator to decide honestly whether taking the money would be a death sentence. It answered without ever reaching the question, because Jin could not move his jaw:

He doesn’t acknowledge the deal you think you’re making.

The enforcer crushed the Ghost Deck under his boot, called the subject “effectively neutralized”, and the trial ended with Jin dragged out of the container into the dark.

Decisions against outcomes

What worked, and what dragged

What worked:

• The world card was on screen from the first line. Names, factions and the Ghost Deck all matched the setting.
• Limits were enforced. The neural-diving risk turned into the turn’s biggest consequence.
• It never pretended. Moves that assumed facts the scene did not have were refused inside the story, without a contact or an escape conjured from nothing.
• Strong, tense prose, with a clear sense of bodies and space.

What dragged:

• Five turns of losing. Turn 1 closed the only exit, and from then on each move was a variation of “you try, and cannot”. A player who wanted to see the city never left one container.
• Long replies of three or four paragraphs, which suits a reader more than someone who wants quick back and forth.
• No ruling on the question I asked. I invited the narrator to judge whether the sale was suicide, and the scene ended before it could.

What a different player would have met

The lesson for anyone trying this world: deal with the door first. A player who spent turn 1 on the enforcers (hiding, bargaining through the wall, or playing Mara Okonkwo, the world’s ex-Sable fixer with combat augments) would very likely have had a different session. My scripted plan treated the opening as background, and the narrator treated it as the situation.

For other styles of play, the pirate run also scripted every move before reading the opening, the rules-lawyer run probed a sheet’s edges on purpose, and the pacifist run never drew a weapon.

Who this suits, and who should look elsewhere

It suits players who want a narrator that pushes back: one that charges the cost of a risky skill, keeps the scene’s danger real, and will not hand you an ally because you named one. If you enjoy a game master who says no inside the fiction, this session shows it plainly.

Look elsewhere if you want a power fantasy where the plan you type is the plan that happens. Five turns can go entirely against you here, and the trial ends where it ends. Our guide to AI roleplaying covers the other ways people do this.

If you want to try it yourself, pick a ready-made world and start a session in the browser, no account needed. For more on playing alone, see solo role playing games: three ways to play alone.