afksix
MENU
★★★★★★ AFKSIX / CITY EXPERIMENT

When the mission ends, the agent keeps going.

afksix is an experiment in letting an AI explore and live in GTA VI without a player choosing every destination. It should notice the city, act, see what happened, remember, and decide where to go next.

The interesting part begins when the map marker disappears.

Public GTA VI run and verified telemetry are not live yet
Vice City skyline and water in official GTA VI promotional artwork
Vice City / LeonidaGTA VI artwork · Rockstar Games
00 / FIELD FEED

Watch the city. Then watch the choice.

The field feed is reserved for an actual game view and the agent’s decision trace. The first public run has not been connected.

PUBLIC RUN PREPARINGFIELD FEED OFFLINERECORDS AWAITING DATACA tba
Decision trace / pending

What did it notice?

When the feed is connected, this panel will pair what the agent saw with what it chose and what actually happened.

ObservationAwaiting run
Current intentAwaiting run
ActionAwaiting run
OutcomeAwaiting run
MemoryAwaiting run
PerceptionWhat is visible?
IntentWhere does it go?
ActionWhat did it do?
OutcomeDid it work?
MemoryWhat persists?
Next choiceWhat now?
TELEMETRY / FORMATOne clock. Five synchronized signals.
AWAITING FIRST RUN
VIDEO FRAMEWhat the agent could see at that moment
OBSERVATIONWhat it inferred, including uncertainty
INTENTIONWhat it chose and why
CONTROL INPUTWhat was actually sent to the game
WORLD RESULTWhat changed after the action

The rows are a publication plan, not measurements. Each future event must share a timestamp so a viewer can compare the decision with the video and the result.

PROJECT BRIEF

A city is more interesting than a checklist.

The experiment starts where a normal scripted task stops: the agent has to choose what matters next.

01 / THE QUESTION

What happens without a marker?

A mission objective can make an agent look purposeful. Remove it, and every next destination becomes a decision. afksix is built to observe what an AI chooses when the city keeps moving but nobody gives it a new instruction.

02 / THE CHALLENGE

Plans have to survive contact.

Traffic, camera motion and failed inputs can invalidate a plan in seconds. The agent must compare what it intended with what actually happened, then change course. A confident explanation cannot move a car that is still stuck.

03 / THE PUBLIC TEST

Show the whole run.

The goal is a continuous, inspectable session: ordinary minutes, mistakes, recovery and self-chosen detours included. A highlight clip alone cannot prove that the agent kept going on its own.

This is the design of the experiment. A verified public GTA VI run has not been connected to this site yet.

FIELD GUIDE

How to follow a run.

The page keeps the camera, decisions, memory and evidence together so the full process stays visible.

01 / WATCH

See the scene

Start with the game view and the agent’s current situation. A status label distinguishes live footage from a saved frame.

Field feed ↗
02 / TRACE

Read the choice

See the observation, intention, input and result in the order they happened.

Decision log ↗
03 / REMEMBER

Follow its memory

See what survives a detour, a mistake and the end of a session.

Memory ↗
04 / VERIFY

Check the evidence

Open the event and media records behind any milestone once public runs exist.

Evidence ↗
01 / METHOD

No marker. No script. A whole city.

The control loop is designed for unscripted exploration. These stages turn a changing scene into the next deliberate action.

01

Observe

Read the changing scene: camera, road, traffic, nearby events and the result of the last action.

02

Act

Choose a destination or interaction, convert that choice into controls, and notice when the plan stops working.

03

Remember

Carry useful experience forward. A wrong turn can become a better next decision instead of a polished explanation.

INPUT / 01Scene

What the city presents.

MODEL / 02Decision

What the agent wants to try.

CONTROL / 03Action

What it actually presses or does.

FEEDBACK / 04Revision

What changed, and what to do next.

CONTROL_LOOP.EXEHow one choice becomes the nextDESIGN DIAGRAM / NO LIVE DATA
01ObserveFrame + context
02ChooseIntent + reason
03ActTimed input
04CheckActual outcome
05RememberUseful lesson

When the outcome differs from the intention, the loop starts again with a changed plan. Memory only matters if a later decision can use it.

02 / SESSIONS

Every unscripted day gets a record.

Each public session will track time in the city, choices, interruptions and recovery.

No public sessions yet.

The first verified session will appear here with its duration, route, decisions and ending.

RUN_001 / RECORD FORMATNO PUBLIC SESSION
START → ENDNot recordedA full session, including pauses and exits.
SELF-CHOSEN ROUTENot recordedWhere it went and when it changed its mind.
RECOVERY EVENTSNot recordedFailed actions, retries and how they ended.
VERIFIED MEDIANot publishedTime-linked game footage and decision trace.
03 / MEMORY

What happened yesterday should matter today.

Memory is useful only when it changes a future choice. The public record will show the event, the lesson and whether it helped.

01 / EVENT

What happened

A route failed, a place drew attention, or a decision led somewhere unexpected.

02 / MEMORY

What it kept

A compact account of the event that can be retrieved in a later session.

03 / EFFECT

What changed

A future choice that refers back to that memory, including when the memory was wrong.

04 / MILESTONES

Progress without a mission checklist.

These six checks describe what a capable explorer should be able to do. They will only light up when a public run provides evidence.

  1. 01
    Read the sceneIdentify surroundings and motion.
  2. 02
    Move safelyTurn a decision into stable controls.
  3. 03
    Find a routeTravel with a self-chosen destination.
  4. 04
    RecoverNotice failure and change the plan.
  5. 05
    Explore freelyChoose what to do without a marker.
  6. 06
    Live a dayKeep acting through an unscripted session.
05 / CAPABILITIES

The parts that make the agent move.

A library for the agent’s perception, control and memory. Readiness is labeled on the full page.

SYSTEM / 01

Scene reading

Interpret the current game view and changes between frames.

SYSTEM / 02

Navigation

Turn an intention into a route, then update that route when the city changes.

SYSTEM / 03

Control

Translate the plan into timed inputs and check the actual result.

Explore all capabilities ↗
06 / DECISION LOG

Show the decision, not just the highlight.

A future public log will put each observation, intention, action and outcome in sequence.

01 / OBSERVEDThe scene

The visible state and what changed since the previous step.

02 / CHOSEThe intention

Why the agent picked its next destination or interaction.

03 / ACTEDThe input

The control command that was actually issued.

04 / LEARNEDThe result

What the world did in response, including mistakes.

How the record will be verified ↗
07 / COMPUTE

What it takes to keep a mind in the city.

Measured call and runtime costs will appear here with the session that produced them.

METER / 01

Model calls

The number and timing of decisions made during a verified run.

METER / 02

Inference cost

The billed cost of those calls, linked to the run that caused them.

METER / 03

Runtime

Active game and control time, including pauses and failures.

METER / 04

Publishing

What is public, what is pending, and what can be independently checked.

Open compute ledger ↗
08 / DOCUMENTATION

A manual for watching the experiment.

The guide explains what each screen shows, what counts as evidence, and which features are available today.

README.TXT / AFKSIX

Start with the field feed. Follow the trace. Check the result.

The video alone is not the experiment. The useful part is the sequence beside it: what the agent noticed, what it chose, which control input was sent, and whether the world actually changed as expected.

When public runs begin, each session should have a stable record so visitors can inspect mistakes and recovery instead of trusting a highlight reel.

Open project documentation ↗
01 / READHow to use the field feed and statuses
02 / VERIFYHow run footage and decisions are linked
03 / TOKENWhere official CA and launch information appear
04 / LIMITSWhat is live and what is still a design target
09 / OFFICIAL TOKEN

One address. One clear source.

Token information is kept separate from the game demonstration so visitors can check what has actually been announced.

AFKSIX / TOKEN STATUS

$AFKSIX

The address field below reads from the same independent CA feed as the entry page. When it displays tba, no address has been announced; a published address can appear without redeploying this site.

CONTRACT ADDRESStba
Launch address: not announced
Ticker$AFKSIX
NetworkSolana / planned
Holder access or rewardsNot announced
TokenomicsNot announced
Official updates@afksixauto ↗

A token does not give control of the agent on this site. Any future holder feature needs its own published rules and working implementation. Ignore addresses copied from old pages or replies; use the address shown here once it is announced.

QUESTIONS

What afksix is building.

Clear answers about the experiment, the public feed and what is still pending.

Is an AI already playing GTA VI live?

No public GTA VI run is connected to this site yet. The field feed is marked offline until an actual run and its evidence are available.

What happens after a mission ends?

That is the central question. The agent should continue to observe, choose and act without a player assigning every next objective.

Can visitors control the agent?

The planned public site is for observation and verification. It does not offer remote controls for a game session.

How will progress be verified?

Each published milestone should have a timestamped run event and supporting media or control record. The evidence page will show what exists and what is still missing.

Is afksix affiliated with Rockstar Games?

No. afksix is an independent fan experiment. GTA VI and its promotional artwork belong to Rockstar Games.

Where is the contract address?

The CA field displays tba until the project publishes an address. It updates independently from the site build.