How to Set Up a Free, Uncensored AI Companion Locally (No GPU)
Written by Clement
You want a private AI companion that doesn't refuse normal conversation and doesn't cost anything. You're on the right tuto. This installs a completely local, uncensored chat model on your own computer in about ten minutes, no GPU required.
Don't want to install anything?
Try GenLovers' hosted local-chat sandbox first, no setup, runs on our own machine, to see what a companion chat feels like before you commit to installing your own.
Why local and why "abliterated"
Local means 100% privacy. Nobody can see what you're doing. Nobody can judge it, store it, or reuse it. It also means you actually learn how a local LLM works, which is worth something on its own, for fun or for work.
A normal instruction-tuned model is trained to refuse certain requests. An abliterated version has that refusal behavior removed directly from its weights, not prompted around. I tested the same base model without abliteration first, and it flat-out refused requests the abliterated version handles without issue. That's the difference in practice, not just in theory.
One-click scripts
Get Ollama and the abliterated model running. Three ways to do it, all ending up in the exact same place, an Ollama server on localhost:11434 with huihui-qwen3.5-0.8b-abliterated pulled. The scripts exist to save you time and mental energy: you don't have to research anything, I already did it, you just click and it downloads. If you'd rather not run a script you didn't write yourself, that's a fair concern, so "do it by hand" below is right there too, and it's not much slower.
Windows: one-click script
Downloads and runs a PowerShell script that installs Ollama, pulls the model, and verifies it responds. No commands to type.
Mac / Linux: one-click script
Same automation as a shell script: installs Ollama if missing, pulls the model, and sends a test prompt to confirm it works.
Talk to it
- 1
Chat directly in the terminal
Run ollama run huihui-qwen3.5-0.8b-abliterated, type a message, press Enter. Type /bye to exit.
This is the raw model. A companion needs a persona on top.
The terminal works, but I'm not convinced it's the best way to talk to it. It's not very enjoyable, it doesn't keep memory between messages the way a normal chat app does, and it's not how anyone actually expects to use an LLM day to day.
What you just installed is also a blank chat model: no name, no personality, no character. Turning it into an actual companion, with a name, a consistent personality, and a way of speaking, is a prompting problem, not an installation problem.
So I built a small self-hostable chat page instead: a normal, easy UI, running entirely in your browser, talking straight to your own local Ollama, and never sending anything anywhere else. Set a name, an avatar, and a personality prompt once, and it remembers your companion the next time you open it. If you try it and want something changed, tell me on Discord and I'll either make the change or walk you through modifying it yourself.
Get new guides by email
One email when we publish new guides and model breakdowns. No spam, unsubscribe anytime.
Want something changed, or stuck anywhere in this guide?
- 2
Download the companion chat page
A single offline HTML file. No build step, no server, no account.
Download the companion chat page - 3
Open it in your browser
Double-click the downloaded file. It opens to a setup screen: name, avatar image URL, personality prompt, and the model tag you pulled above.
- 4
Start chatting
Fill in the setup screen and hit Start chatting. Your companion's config is saved in your browser for next time.
What you need
I run this exact setup on a nearly 8-year-old Acer laptop, one of the cheap ones, 8GB of RAM, no GPU, and it still runs well. If you don't own a GPU, this is still a good pick: it puts out a real number of tokens per second even on CPU alone. The tradeoff is accuracy, not speed: a small local model like this won't answer as well as a much bigger hosted one, but it's yours, it's private, and it's uncensored.
| OS | Windows, Mac, or Linux |
|---|---|
| GPU | None needed. Tested on an 8-year-old laptop with 8GB RAM and no GPU. |
| Disk space | Ollama itself plus about 528 MB for the model file |
Troubleshooting
| ollama pull fails partway through | Retry the pull. If it keeps failing on the same file, the setup script's fallback handles it automatically |
|---|---|
| Replies never finish, CPU pinned at 100% | The model is a reasoning model and needs think: false set on the request; the companion chat page already sends this, a raw ollama run in the terminal does not |
| Wondering how fast replies will be | Try the local-chat sandbox linked at the top of this page first to get a feel for it. Your own machine will likely be faster: it's probably better hardware than my old Acer, and you skip the sandbox's server queue and cold start entirely |
| Companion chat page says "Ollama unreachable" | Make sure Ollama is running in the background, and that the Ollama server URL field matches where it's listening (http://localhost:11434 by default) |
| ollama: command not found after install | Restart the terminal so it picks up the updated PATH, or reinstall |
| Stuck during the one-click setup | Message the Discord linked above and I'll help you sort it out |
Frequently asked questions
- Do I need a GPU?
- No. This setup is CPU-only by design, using a small model so replies stay usable on ordinary laptop hardware.
- Is this actually private?
- Yes. Once the model is downloaded, nothing is sent anywhere. The model runs as a local process, and the companion chat page talks to it directly over your own machine's loopback address.
- What does "uncensored" mean here exactly?
- The abliterated model has had its built-in refusal behavior removed from the weights, so it engages with companion-style conversation a stock safety-tuned model would deflect. It's still a general chat model, not a tool for generating explicit media.
- Why does GenLovers point to its own companion chat page instead of just the raw Ollama setup?
- The raw setup gives you a blank model with no character. The companion chat page adds a name, an avatar, and a persistent personality prompt on top, so you get an actual companion instead of building that groundwork yourself in the terminal.
Keep reading
How to Set Up a Local NSFW Image Model: Full Tutorial
Install an adult image-generation workflow on a rented cloud GPU: RunPod, ComfyUI, a real workflow file, and every step in between, screenshot by screenshot.
How to run MiniMax H3 locally in ComfyUI
Install MiniMax H3 in ComfyUI: the right FP8/INT8 files for your GPU, the license restriction to check first, and all three workflows, step by step.
What is a LoRA and when do you need one
A plain-language explanation of LoRAs for AI image generation: what they add to a base model, why they only work on open-weight models, and what weight to load one at.
Get new guides by email
One email when we publish new guides and model breakdowns. No spam, unsubscribe anytime.