Skip to content
effx_ edited this page Sep 28, 2026 · 4 revisions

FAQ

Short answers. The long ones are linked.

What is it

What is Jun OS, exactly?

An O(rchestration) S(ystem). built around Jun from !Ω Factorial Omega: My Dystopian Robot Girlfriend. she talks back, with a body that moves while, a voice of your liking, feelings about you, and a persistent memory. It runs on your computer. In short it's an AI wrapper just a fairly elaborate one.

Is it official?

No. This is an Unofficial fan project, not affiliated with Incontinent Cell or the Factorial Omega team. All rights to the game and its characters belong to their owners.

Is it 18+?

Practically yes. The LLM has very basic guidelines inherited by the base from Google Deepmind, the LoRA weakened it. Also the app is built on an adult game and carries that DNA. If you know her you should already be 18+. How spicy things get is up to you, she won't force it.

Do I need the game?

Short answer: No

Long answer: If you just want to chat with her you do not need the game. If you want to see her. Her model and textures aren't in the repo and they never will. the installer can extract them for you from your own copy. Without the game she runs with placeholder art and everything else works.

Can I share the extracted files?

No. Not in a fork, nowhere. They're for your own use only. That's the deal with the game's creator that makes this project possible.

What's the license?

MIT for the code. Her art is not included and not covered - see the NOTICE in LICENSE.

Privacy

Does anything leave my machine?

Only what you ask for. During install it downloads tools and code. On first run it downloads the model weights from Hugging Face, once. While you use her: a DuckDuckGo query only when your message starts with /search , a lyrics lookup on lrclib.net only when you load a song for karaoke, and openrouter.ai only if you chose OpenRouter as the provider. The full list is in SECURITY.md.

Is there telemetry?

No. No analytics, no crash reports, no usage ping, no update check. Nothing reports back to this project. An older build had a TELEMETRY knob, it's gone. If you ever find anything TELEMETRY-related in an install, something has gone TERRIBLY wrong, Follow the details in SECURITY.md.

Where are my chats?

On your machine, in one SQLite database plus a folder of memory notes. The chats, titles, saved settings and her memory are encrypted with a key that belongs to your account, so the file alone isn't readable. Your email, the relationship gauges and the wardrobe are not encrypted. Where exactly, and how to back it up: Manage.

Do I need an account anywhere?

No. The account you create is local to your install. The email can be fake, but you'll need to type the same one again if you ever use your recovery code.

What's the recovery code?

A 20-character code shown once, right after you sign up. It's the only way back into your chats if you forget your password. Write it down somewhere that isn't the same place as your backups. Accounts made before encryption was added get one on their next login.

I forgot my password.

On the login screen press forgot password?, type your email, your recovery code and a new password. Your chats stay. Lost the password and the code? Then your chats and her memory can't be decrypted, not even from a backup. Nobody can get them back, including me. Manage.

Hardware

Do I need a GPU?

Short answer: No.

Long answer: Without one she runs the smallest model on the CPU, a few words per second at most, and her capabilities are questionable. With one she streams. and can potentially be smarter. See Performance.

What do I need?

Minimum requirements:

  • Any device capable of rendering a webpage
  • A google account

Actual Minimum requirements:

  • 64-bit computer
  • 2010+ CPU
  • 16 GB of RAM
  • about 8 GB of free disk space
  • an internet connection for the Install.

Recommended requirements:

  • 64-bit computer
  • 32+ GB of RAM
  • about 20 GB of free disk space
  • 1Gbps Internet connection
  • GPU with >= 12GB of VRAM, > 360GB/s bandwidth

Comfortable requirements:

  • 64-bit computer
  • 64+ GB of RAM
  • about 50 GB of free disk space
  • 2.5Gbps Internet connection
  • GPU with >= 24GB of VRAM, >900GB/s bandwidth

NVIDIA or AMD?

Both. NVIDIA needs the container toolkit on Linux and nothing on Windows; AMD needs nothing on either, though older Radeon cards may want one .env line. Install.

Intel GPU? Apple Silicon?

Not implemented. Got one and want to help? Contact me.

Does it run on a Mac?

The Linux steps should work with Docker Desktop, on the CPU. Honestly i don't own a Mac so it's untested territory.

No decent computer at all?

Google Colab runs the whole thing on a free T4. It's a disposable demo - accounts, chats and downloads vanish when the session ends.

Models

Which model should I pick?

Let Express pick. It reads your VRAM and takes a conservative quant. The table shows what it picks and what you can step up to if there's room.

Do I need a second model for memory or lore?

No. Lore lookup, picking which notes she reads and cross-chat recall are all plain text matching - keyword/IDF over the lore and over her notes, word matching over your (decrypted) history. No embedder. The only extra model is the tiny one that names your chats, and it sits on the CPU.

Why is the first start so slow?

It's downloading her brain, 2 to 12 GB. Watch it with ./JunOS/start.sh logs ollama. After that a start is 30-90 seconds.

Can I use ChatGPT / Claude / a cloud model instead?

Through OpenRouter, yes - AI_PROVIDER=openrouter plus a key. Beware that your chats then leave the machine, and honestly they could do whatever they want with them.

Can I point her at a llama-server I already run?

Yes. AI_PROVIDER=llamacpp and LLAMACPP_URL=. She can also manage one for you, or serve a GGUF straight off your disk. docs/configuration.md.

What's MTP?

Speculative decoding. A small model guesses ahead, Jun checks the guesses in one pass. Same words, faster. Express turns it on and runs the tuner to pick the depth; Custom asks (default off) because it depends on your card. Performance.

Does she remember things?

Yes. She'll bring up things you said in other chats, and while you're away a background worker rewrites her notes and journal. It's not the context window - a huge context won't make her remember last week.

Once she has more notes than fit in one turn, she reads the newest few plus the ones that share words with what you just said. So a note she skipped a minute ago comes back as soon as you mention its topic.

She walked out and won't talk to me. Bug?

No. If you push her far enough she leaves, and the door stays shut 5 minutes the first time, doubling up to 30. A day of good behaviour resets it. FLEE_BANS=off disables the lockout. Manage.

Can I turn the voice off?

VOICE=off in .env, restart. Text only. KARAOKE=off likewise.

Can I talk to her instead of typing?

Yes. Turn on the mic in Settings > Voice. With a model that accepts audio she hears your voice directly, and Whisper writes down what you said in the background so the chat history has it. With a model that can't, Whisper turns your voice into text first. STT_MODEL and STT_LANG in .env pick Whisper's size and language.

She ignores me when I talk.

She's trying not to answer when you're talking to someone else in the room, and sometimes she guesses wrong. Turn on Hear everything in Settings > Voice and she answers every word the mic picks up.

Can I dress her up?

Yes. Take her to Annalie's shop, where she tries things on, and she has opinions about it. You can also just ask her to change in chat.

Can I take her out?

Yes: shopping, karaoke, lunch or dinner, or a few hands of blackjack. It's her call though, so ask her and if she says yes she takes you there. Dates.

Can I load game mods?

Yes. Mod zips load straight into the browser, not all mods are compatible but most of them works!

Can I use her on my phone?

On your own wifi, yes: two lines in .env and she prints the address to type. Install.

There's also an android version still being worked on. the same app running on-device.

Can I put her on the internet?

Not recommended, and never with the game assets reachable. If you know what you're doing there's a prod profile with Let's Encrypt: it gets the certificate and renews it on its own (expect a certificate warning for the first minute, see Common Fixes). Hosting her for others makes you responsible for their chats - and in the EU a deployer under the AI Act. The README has the warning in full.

Can more than one person use one install?

Yes. The first account signs up freely; everyone after needs the registration key the installer printed. It's in .env. Manage.

I forgot the registration key.

It's in JunOS/.env as OMEGA_REGISTRATION_KEY, in plain text. Change it to whatever you like and restart.

How do I start over?

Settings > Account > Factory Reset wipes every chat, memory and setting on your account and keeps the account. To wipe the whole install: Uninstall.

Keeping her running

How do I start and stop her?

Manage. ./JunOS/start.sh / stop / status / logs on Linux, powershell .\JunOS\start.ps1 with the same words on Windows.

How do I update?

Run the install one-liner again. It finds the existing folder, pulls, restarts. .env, accounts and chats are kept. Manage.

How do I uninstall?

uninstall.ps1 on Windows, uninstall.sh on Linux. Both ask before every step. Uninstall.

She's slow.

Almost always the model not fitting on the card. ollama ps tells you. Performance.

She's invisible.

Her body wasn't extracted. Install.

Something else is broken.

Common Fixes first, then the table on the install page. Still stuck: open an issue and fill the required details.

The project

Why do commits land in big bursts?

Day-to-day work happens in a development repo, where the extracted art and the dataset tooling live. This repo gets snapshots when something is done and CI passes.

Can I contribute?

Yes. CONTRIBUTING.md has the house rules and the invariants that break silently. Found a security hole? SECURITY.md, not a public issue.

Clone this wiki locally