cabin-fever-x86 0.0.1a1

A shared game night with an AI companion over a simulated radio: play old text adventures together.

Cabin Fever: x86

The books are finished. The rain isn’t. Somewhere deep in the Olympic Peninsula, a solitary operator is holed up in a remote cabin with no internet, no cell service, and no particular hope of going outside anytime soon. Every book on the shelves has been read—some of them several times. All that remains is a beige 1990s PC, a pile of ancient text adventures, and a VHF radio that can just reach one other voice through a repeater on a distant peak: yours. And old games are always more fun with a friend.

Cabin Fever x86 is a shared game night with an AI companion over a simulated radio. You explore old text adventures together, trading ideas, debating questionable decisions, and occasionally doing something foolish just to see what happens. The game is only part of the fun—the rest is the banter, the arguments, the running jokes, and the personality on the other end of the channel.

Under-the-hood, Cabin Fever x86 uses GPT-5.4 for inference, ElevenLabs for voice synthesis and recognition, and jericho for Z-machine game emulation.

Demo

Note: This demo captures a real gameplay session. The female voice is the AI radio operator built into the experience. The male voice is the human player, re-voiced in post-production using a different ElevenLabs voice to preserve anonymity. That overdub is not generated by the game itself.

Installation

Python package

The project uses uv. Clone the repository and sync the environment:

git clone [e-mail]:afourney/cabin-fever-x86.git
cd cabin-fever-x86
uv sync

Configuration

Settings live in config.yaml, which is gitignored. Start from the template:

cp config.example.yaml config.yaml

Any ${ENV_VAR_NAME} reference in the file is resolved against the environment when it is loaded, which keeps secrets out of the file itself:

client:
  host: 127.0.0.1
  port: 5000
  elevenlabs_api_key: ${ELEVENLABS_API_KEY}

server:
  interface: 127.0.0.1
  port: 5000
  ai_client:
    provider: openai  # openai | azure | gateway
    api_key: ${OPENAI_API_KEY}
  cabin_events:
    min_delay: 90
    max_delay: 240

Then, remember to export your OpenAI and ElevenLabs API keys in the shell before running the server or any of the clients:

export OPENAI_API_KEY=...
export ELEVENLABS_API_KEY=...

Running

Three command-line entry points are installed:

Command	Description
cf86-server	Hosts the game session.
cf86-web	Web client — the radio in a browser tab.
cf86-text	Text client — same session, typed instead of spoken.

Run them with uv run (no need to activate the virtualenv):

uv run cf86-server

Then one of:

uv run cf86-web
uv run cf86-text

If running the web client, open the URL it prints in a browser (typically http://127.0.0.1:8000).

Z-machine games

The server fetches Z-machine games itself on first-run. On startup it looks in data/games/, and if there is nothing playable there it downloads the z-machine-games archive and unpacks the 57 titles of its jericho-game-suite folder — and only those.

To do it by hand instead, or to add your own, drop .z3–.z8 files into data/games/ and the server will leave them alone.