Desktop apps, model runners, image generation, offline transcription, private document chat and coding assistants. Each tool is described in our own words, with a link to its official site.
What a featured listing costs · from $199 a month
LM Studio is a desktop app for running open language models on your own computer. You search for a model inside the app, pick a file size that fits your…
Jan is an open-source desktop app for chatting with language models. It downloads open models and runs them on your own hardware, and it can also connect to hosted models…
GPT4All is a desktop app from Nomic AI for running language models on everyday computers, including ones without a strong graphics card. It offers a list of downloadable models sized…
Open WebUI is a chat interface you host yourself, usually on the same machine as a local model runner such as Ollama. You open it in a browser and get…
Ollama runs open language models from the command line. You pull a model by name, run it, and Ollama handles the download, the quantized file and the settings needed to…
llama.cpp is an open-source project for running language models in plain C and C++. It introduced and maintains the GGUF file format, which stores a model and its settings in…
Tools come from submissions and from our own reading. A tool must be live and available to download or install today.
We open the official website or code repository and confirm the tool runs models locally the way our listing says.
Each description is a short account, in our own words, of what the tool does and who it suits.
If we can't confirm a detail, such as Linux support or a local API, the listing says "not enough info".
A tool is listed once we have confirmed that its official site or repository is live and that it does what our description says. Each tool sits in the one category that fits its main job, even when it covers several. We don't publish speed scores, rankings or prices, because results depend on your hardware and a score would be our opinion presented as a fact.
Running a model locally keeps your prompts on your machine, but the model itself still comes with a license. We link to official sources, and we ask you to read each model's terms before you use it for work.
Makers can submit or claim a listing. A person checks every request. Published by Zwings LLC.
Start with the job. Chatting with a general model, transcribing interviews, generating images and searching your own files are different tasks with different tools, and a good fit for one is rarely a good fit for another. Pick the category first, then compare the tools inside it.
Next, look honestly at your hardware, because it decides more than any feature list. A language model has to fit in memory to run well. On a Windows or Linux PC with a graphics card, the amount of video memory on the card is the number that matters most. On an Apple silicon Mac, the CPU and GPU share one pool of memory, so total RAM sets the limit. Laptops without a dedicated GPU can still run small models on the CPU, only more slowly.
Quantized files are how most people fit models onto ordinary machines. A quantized model stores its weights with fewer bits, which cuts the memory it needs a great deal at the cost of some quality. Most local apps offer several sizes of the same model. A sensible approach is to start with a small one, see how it answers, and move up while your machine stays responsive.
Decide how much setup you want. Desktop apps install like any other program and hide the details. Command-line runners and web interfaces you host yourself take more effort, but they give you a local server that other apps can share. If you are not sure, a desktop app is an easy start, and many of them can run that server too.
Check what actually stays on your machine. Some tools are fully offline once a model is downloaded. Others mix local models with hosted ones, collect usage data or check for updates. Read the privacy settings, and if offline use matters to you, test the tool with your network turned off.
Read the model license as well as the tool license. The app may be open source while the model you run with it limits commercial use, requires attribution or forbids certain uses. The license sits on the model's page, usually on a model hub.
Be careful with add-ons. Image interfaces and some chat apps support community extensions and custom nodes, and each one is code you are choosing to run. Prefer well-known sources, and prefer safetensors files over older formats that can run code when loaded.
Finally, keep expectations realistic. Local models have improved quickly, but the largest hosted models still do better on hard reasoning. For private notes, drafts, transcription and coding help, a local setup is often enough, and your data never leaves the room.
No. We describe what each tool does and who it suits, in our own words, and link to the official site or repository. There are no scores or rankings. Paid placements are marked Sponsored and do not change our descriptions.
It depends on the model. Small language models run on most recent laptops, even on the CPU. Larger models need more memory: video memory on a PC graphics card, or total RAM on an Apple silicon Mac. Image generation is easier with a dedicated GPU. Start small and move up.
The model itself runs on your machine, so prompts and files don't have to leave it. The app around it may still check for updates, collect usage data or offer hosted models. Read each tool's privacy settings, and test with the network off if that matters to you.
Sometimes. Every model has its own license, separate from the license of the app you run it in. Some allow commercial use, some limit it, and some require you to accept terms first. Read the license on the model's page before you build on it.
Speed depends on your exact hardware, drivers, model size, quantization and settings, and it changes with every release. A number measured on our machine would mislead you on yours. The hardware utilities category lists tools for measuring your own setup.
It means we could not confirm that detail, such as Linux support or a local API, from the tool's own site or repository. We leave it marked that way instead of guessing, and update it when the maker confirms.
Use the submission form, or email us. We check every request by hand, write the description ourselves, and correct listings when a tool changes, moves or stops being maintained.