Meterless Experiments · 01

The city
talks back.
And remembers.

Walk up to anyone in a photoreal city and talk out loud. They answer in their own voice, remember you next time, gossip about you to their friends, and might join your crew. Every line runs on your own Meterless Runtime, so nobody gets a bill.

Free to play$0 per conversationRuns on your own AI
  • No cloud bill
  • Memory that lasts
  • Mods are HTML files
Greenwood · live captureNo cuts to the AI
Recorded in play · Powered by Meterless Nightshift
01 / The trailer

Two minutes
in Greenwood.

Everything here was captured in one sitting on one PC, with the Meterless Runtime on a second machine on the same network. The voices, replies, choices and memories are real output, not scripted.

02 / What you do

Talk is the
whole game.

Conversation is the core mechanic. What you say moves trust; trust brings people into your crew; your crew earns money, influence and rumors; and the whole city sees you through those numbers.

03 / The phone

Press L3.
Your city, in apps.

Text anyone you've met and they text back with their memory of you. Read the city's front page, bet on fight night, book a crashpad, buy a car, follow the crews. Adding an app means dropping one HTML file in a folder.

04 / On the move

Phone up.
City moving.

The phone never stops the game. Read your messages crossing the street, check the World Model as traffic cuts past, scroll After Dark on a drone flight over the harbor, book a crashpad while you jog the pier. The people you pass still answer when you speak.

  • On foot · walk or jog
  • Drone · fast travel
  • Ride-alongs · in traffic
  • Subway · three lines
Phone up · live captureWalking · flying · riding
Recorded in play, the phone open the whole way · nothing staged
05 / The city

Greenwood,
as it might have been.

An alternate history where the 1917 and 1921 mobs were stopped and the 1923 Restitution Act paid what was owed. A century later Greenwood is a Chicago-sized city, and its people have opinions about everything, starting with Meterless Nightshift.

06 / Under the hood

Small model.
Long memory.

An 8B model on a home runtime can feel like a frontier model when the systems around it are built for one GPU. These are the Meterless engines doing that work inside the city.

Markovian

The 100th line costs what the first did.

Each person carries a bounded state (completed, remaining, decisions, context) plus the last four turns, never the whole transcript.

Engine from github.com/meterless
H-MEM

They remember last week.

Short-term, working and long-term memory with a sleep cycle that consolidates old episodes into chapters. Recall comes back as context on every line.

An hour of play is a memory day
World Model

Every fact has a source.

Who said what, who joined whom, who was sent where. An event log with provenance and relationships that change over time. The phone shows you all of it.

Nothing about you is hidden from you
Think-ahead

Answers ready before you ask.

While you walk, Meterless writes the lines you might say to the people near you, their answer to each, and the voice clip. Ready lines show green and play at once.

Cancelled the moment you talk
Multi-ask · multi-reply

Ask three things at once.

Questions get split and each one answered. Replies come in up to three beats: an answer, a second thought, a question back. Nothing you say gets dropped.

After Gaia's Multitask Mode
One runtime

Your hotspot, every app.

The city reaches your Meterless Runtime over one OpenAI-compatible URL. The same hotspot serves your other Meterless apps. Text generation never falls back to a metered cloud.

Local hearing fallback: Whisper on your GPU
07 / Why it's free

No meter on
a conversation.

Hosted AI characters bill per line, so a hit game is a cost disaster and the characters go quiet when the vendor does. On your own runtime the cost of talking is electricity.

~700NPC turns in an active hour of play
66%answered instantly from Meterless-written pools
~0.5slocal Whisper hearing per line
$0per conversation, forever
On your Meterless Runtime

≈ $0.02–$0.06 an hour

Electricity for the runtime box under load. Local models are not metered.

Same workload on metered cloud APIs

≈ $0.46–$6.60 an hour

About 445 model calls, 22,500 characters of speech and 3 minutes of transcription per active hour, at mid-2026 list prices.

Estimates from our own play logs (2,347 NPC turns, 2 Oct 2026) and public list prices. Cloud prices change often; yours will differ.

08 / Controls

Keyboard or
controller.

No microphone? Every conversation also works by picking one of four lines Meterless writes for the moment: friendly, curious, bold, or a change of subject.

Do thisKeyboardController
Talk out loudHold CHold talk
Four things to sayTap C, pick 1–4R3, D-pad picks, LB rerolls
Type anythingF8n/a
Your crewF6RB in the options
PhoneF5L3
Map, fast travel, subwayMPhone › Maps
Ride along with a carM next to itRT next to it
Go insideF9X in the options
09 / What you need

A PC and
a runtime.

Windows 10 or 11 with a modern NVIDIA GPU and a microphone (optional), plus a Meterless Runtime: on the same PC, or on a second machine with hotspot mode on. Paste its connect link once and the city finds it.

Meterless City is an experiment from Meterless. It runs on Epic Games' City Sample and gives new life to the Smart NPCs demo, whose original hosted service shut down. It is not affiliated with or endorsed by Epic Games or Replica Studios. Characters, places and events in Greenwood are fictional; real musicians, teams, fighters and labels appear only by name.

When the work deserves its own shift

Give your city
the whole night.

Meterless City runs on Meterless Nightshift, the flagship Meterless experience: one AI hotspot for every app you use, local and unmetered.