Common issues
- "Model not found" — you need to download the model first before you can use it.
- App won't start — another program may already be using the same connection port. Restarting your computer usually fixes this.
Most problems are common, and most have a simple fix
Running AI on your own computer, without relying on the internet, gives you privacy and control. But things don't always work the first time. The good news: almost every issue people run into falls into one of a handful of categories, and none of them require deep technical knowledge to solve.
Beginner tip: if you're just getting started, an all-in-one app like Ollama or LM Studio will save you from most of the problems on this page. Tools that require Python and terminal commands are more likely to break for newcomers.
"I can't even get it installed"
If a tool like Ollama, LM Studio, or GPT4All fails right at the start, it's usually one of these three things:
No compatible graphics card, or outdated drivers
This doesn't stop you from using local AI — it just runs slower, using your processor (CPU) instead of your graphics card (GPU). It's still worth updating your graphics card drivers, since that alone fixes a lot of odd errors.
Not enough disk space
AI model files can range from 2 GB to over 40 GB. Check that you actually have enough free space before downloading one.
Antivirus or firewall blocking the install
Some security software mistakenly flags local AI installers. Check whether the file was quarantined before assuming the download itself is broken.
Beginner tip: start with an all-in-one app like Ollama or LM Studio. They come ready to use, with no need to install Python packages or run commands separately — which is where most installation errors happen.
"It's really slow" or "the app keeps crashing"
This almost always comes down to memory — either your computer's regular memory (RAM) or your graphics card's memory (VRAM) not being enough for the model you picked.
- Pick a smaller model. Models are labeled by size, like 1B, 3B, 7B, or 13B — the bigger the number, the more capable but also heavier and slower. On a modest computer, start with a 1B or 3B model.
- Close other programs while using the AI, especially a browser with many open tabs.
- Check your free memory. Simple rule of thumb: if the model file is 4 GB, you want at least 6 GB of free memory for it to run smoothly.
- Responses cutting off mid-sentence? That's usually a response-length limit being hit. Most apps have a setting called something like "max response length" or "max tokens" — just increase that number.
"The AI's answers don't make sense"
Small models make more mistakes
This is normal and expected — a 1B or 3B model is fast and light, but less capable than a larger one. If answer quality matters a lot to you, try a bigger model, if your computer can handle it.
The AI repeats the same sentence over and over
This is usually fixed by adjusting a setting called "temperature." If it's available, try raising it slightly — it controls how much variation the AI allows itself in its answers.
The AI ignores your instructions
Sometimes this happens because the app is using the wrong "chat format" for that specific model. Each model family expects instructions in its own particular format, and using the wrong one can make responses feel broken or off-topic. Check your app's settings for a "prompt template" or "chat format" option matching the model you downloaded.
Quick fixes by tool
A few issues are specific to the app you're using. Here are the most common ones:
Common issues
- Downloaded model doesn't show up in the list — check the folder LM Studio is looking in under settings; it may differ from where the file was saved.
- App freezes loading a large model — lower the GPU usage setting and increase it gradually instead of maxing it out at once.
Common issues (Stable Diffusion and similar)
- "Out of memory" errors — try generating at a lower resolution.
- Images look strange or blurry — check that the correct VAE (an extra model component) is loaded; it sometimes needs to be downloaded separately.
Still stuck?
- Update the app, your graphics card drivers, and Python if you use it — many bugs are already fixed in newer versions.
- Search the exact error message in quotes; someone has likely hit the same issue before.
A simple way to diagnose any problem
When something isn't working, work through these steps in order:
- Restart the app (and your computer, if needed). It sounds obvious, but it resolves a surprising number of cases.
- Read the error message carefully, even if it looks technical. It often already tells you the problem, like "out of memory" or "file not found."
- Test with a small, well-known model. If a small model works fine, the issue is probably that the bigger model is too heavy for your computer — not a broken install.
- Copy the exact error message and search it in quotes. Someone else has very likely run into the same thing.
- Update everything — the app, your graphics card drivers, and Python if you use it.
Still stuck after all that? Consider switching to a simpler, more "closed" app like Ollama or LM Studio, testing with a much smaller model, or running in CPU-only mode to rule out your graphics card as the cause.
Back to top ↑