🌱 The Smallest Uncensored AI Yet — 806 MB

[GrimXLock]

Cerberus Just Got 3 Big Upgrades — Same App, Sharper Brain, Still Free

Chapter 4 in the saga. New file manager, bring-your-own-AI imports, and a pocket-sized 800 MB model.

---

Three things changed in this update — and one of them is genuinely big. A new “file manager” finally shows you what’s eating your disk. You can now drag any open-source AI brain into the app. And there’s a brand-new pocket-sized model that runs on basically any laptop.

Already running Cerberus from earlier? Same install, same key — just pull the update and you’ll see the new panel. Brand new? The links are below. Still 100% free.

---

🗂️ Update 1 — The Model Manager (finally, a real view of your AIs)

Until now, the AI models you downloaded lived as invisible files somewhere on your disk. You knew you had them. You just couldn’t see them. The new Model Manager fixes that.

What it shows you	Why you’ll actually use it
Every model you’ve pulled	One clean list — no hunting in folders
Total disk usage (e.g. “8.94 GB”)	Know exactly what’s eating your storage before you grab a 5 GB model
OLLAMA MODELS tab	The standard ones from Cerberus’s catalog (the “store-bought” AIs)
RAW FILES tab	Custom AI files you imported yourself (your own additions)
Search bar	Find a specific model in two seconds when you’ve got 10

Bridge: Think of it like the Storage settings on your phone — that screen that finally shows you which apps are hogging space and lets you yank the ones you don’t open anymore. Same idea, but for AI brains instead of TikTok and Candy Crush.

Trick: The disk-usage number at the top is your honesty meter. Most people end up hoarding 4–5 models “just in case.” Open the manager once a month, see the GBs you’ve forgotten about, delete the dust collectors.

📦 Update 2 — GGUF Import (bring your own AI brain)

This is the genuinely big one. You’re no longer stuck with just Cerberus’s models.

A GGUF file is the standard “single-file” format for local AI models — picture it as the .mp4 of the AI world. One file = the whole AI’s brain, ready to run. Hugging Face (the GitHub-of-AI website) has thousands of them, free to download.

Where to grab GGUF files	What kind of AI you’ll find
Hugging Face	The biggest catalog — millions of models, every flavor
TheBloke’s collection	The legendary uploader — almost everything popular pre-converted
bartowski’s quants	High-quality recent quantizations of the latest models

How to actually use one:

1. Download a .gguf file from any of those sites
2. Open the Cerberus app → Model Manager → RAW FILES tab
3. Drag the file in (or use the import button)
4. It shows up in your model list. Click. Chat. Done.

Bridge: Imagine a music player that used to only play songs from one store. Now you can drop any MP3 you legally own into it and it just plays. That’s the upgrade — same simplicity, way more options.

Trick: Want a coding-specialist AI? An AI that’s pure roleplay? An AI fine-tuned on legal documents? They all exist as GGUFs. Search Hugging Face for the niche, grab the file, drag it in. Specialist AI for free.

🐣 Update 3 — Gamma3 1B (an uncensored AI that fits in 800 MB)

The new tiny model. 806 MB at the smallest size. Smaller than a single high-res movie. Built on Gemma3 architecture (Google’s open-source AI design — the bones of the brain) with the abliteration treatment applied (= the “I can’t help with that” surgically removed, covered in chapter 2).

Size variant	File size	What it costs you	Best for
Gamma3 Q4_K_M	806 MB	Tiny	Older laptops, low-RAM machines
Gamma3 Q8_0	1.07 GB	Small	Most setups — recommended default
Gamma3 F16	2.0 GB	Modest	Anyone who wants peak quality

What “Q4 / Q8 / F16” means in plain English: Compression levels for the AI’s brain. Q4 = the most compressed (smallest file, slightly less sharp answers). F16 = uncompressed full quality (biggest file, smartest replies). Q8 is the sweet spot. Same idea as choosing 1080p vs 4K for a movie — pick what your hardware can actually play.

Honest catch (straight from the dev): Gamma3 is fast, but it’s still being tuned. “Might have to ask it a couple times to get a response.” It’s a work-in-progress while OP tests new abliteration scripts. Use it for quick stuff; reach for Cerberus 4B v2 or Arbiter when you want a one-shot answer.

Why this matters: Most uncensored AI models start at 4 GB and climb. Gamma3 is the first sub-1 GB uncensored model that actually works — which means your old Chromebook, your dad’s spare laptop, even some phones can now run a private, no-filter AI offline. That’s a category that didn’t exist a month ago.

📥 How to Get the Update (30 seconds)

Already have Cerberus installed?

1. Open the app
2. Hit the PULL button on whichever new model you want (gamma3-1b-abliterated is the smallest)
3. Or drop a .gguf file you already have into the new Model Manager
4. Done

Brand new? Two-stop setup:

Step	Where
Grab the desktop app	access.cerberusai.dev (full install walkthrough in chapter 3)
Browse all available models	llm.cerberusai.dev — direct downloads + total downloads counter (currently almost 5,000)

For terminal nerds (optional power-user move):

The whole thing also has a public API — you can pull models with one curl or wget command, no app required:

wget https://llm.cerberusai.dev/models/gamma3-1b-abliterated/gamma3-1b-abliterated-Q4_K_M.gguf

Resume trick: If a download cuts out, swap wget for wget -c and the same URL — it picks up exactly where it stopped. No starting from zero.

📚 The Cerberus Saga — All the Chapters

If you missed earlier chapters, here’s the reading order:

Chapter	What it covered
Chapter 1	Cerberus 4B — the original 2.6 GB uncensored AI
Chapter 2	v2 — sharper, darker, the daily driver
Chapter 3	The desktop app (full install + setup walkthrough)
Chapter 4	You are here — Model Manager, GGUF import, Gamma3

---

Quick Hits

Want	Do
Smallest possible AI	→ Pull gamma3-1b-abliterated (806 MB)
Daily-driver, balanced	→ Pull cerberus-4b-v2-abliterated (2.6 GB)
Beast mode, 9B brain	→ Pull Arbiter-GL9b (5.8 GB)
Bring your own AI	→ Drag a .gguf into Model Manager → RAW FILES
See what’s eating disk	→ Model Manager → top of the panel
Just browse all models	→ llm.cerberusai.dev

---

Chapter 4 dropped. Your install just got smarter — and it still belongs to you and your hardware only. What are you pulling first?

[Abdullah_Mansoor]

@GrimXLock

Arbiter-GL9b

Is there a Q6 or Q4 version of Arbiter-GL9b?

[tali]

This kind of thing is scary; it can’t be safe for my PC, can it?

[GrimXLock]

its just a hardware probe