What you can build with local AI
Discover what you can build with local AI, check whether your computer can run it, and follow a step-by-step path checked against the official docs. Each card links to a full guide, and its badges show what the Model Finder would set up on a modest machine of each kind.
Written by Jakub Rusinowski · Last updated
12 of 12 use cases
Chat with your PDFs privately
Ask questions about your own PDFs and get answers that cite the page, without uploading the files anywhere.
Uses: AnythingLLM Desktop · Ollama · Open WebUI · LM Studio · Desktop app, Browser
Open the guideModel Finder pick, estimated
- No GPU UsableOn a PC with 16 GB of RAM and no graphics card: usable with Qwen 3.5 4B
- 8 GB GPU SmoothOn a PC with an 8 GB graphics card: smooth with Qwen 3.5 4B
- 24 GB GPU SmoothOn a PC with a 24 GB graphics card: smooth with Gemma 4 26B-A4B
- Mac UsableOn an M1 Mac with 16 GB of memory: usable with Qwen 3.5 9B
Private meeting notes
Turn a meeting recording into a transcript and a summary of decisions and action items on your own computer.
Uses: Meetily · whisper.cpp · Ollama · Desktop app, Terminal
Open the guideModel Finder pick, estimated
- No GPU UsableOn a PC with 16 GB of RAM and no graphics card: usable with Qwen 3.5 4B
- 8 GB GPU SmoothOn a PC with an 8 GB graphics card: smooth with Qwen 3.5 4B
- 24 GB GPU SmoothOn a PC with a 24 GB graphics card: smooth with Gemma 4 26B-A4B
- Mac UsableOn an M1 Mac with 16 GB of memory: usable with Qwen 3.5 9B
Smart-home assistant with Home Assistant and Ollama
Ask your smart home questions in plain language, answered by a model on your own network, with only the devices you choose exposed.
Uses: Home Assistant (Ollama integration) · Ollama · Whisper (Wyoming) · Piper (Wyoming) · Home server, Browser
The Model Finder does not plan this job yet. The guide says what hardware it needs.
Open the guideA coding assistant in your editor
Chat, autocomplete and multi-file edits in VS Code, served by a model on your own machine.
Uses: Ollama · Continue · Code editor
Open the guideModel Finder pick, estimated
- No GPU UsableOn a PC with 16 GB of RAM and no graphics card: usable with Granite 4.2 3B
- 8 GB GPU SmoothOn a PC with an 8 GB graphics card: smooth with Granite 4.2 3B
- 24 GB GPU SmoothOn a PC with a 24 GB graphics card: smooth with Qwen3-Coder 30B-A3B (MoE)
- Mac UsableOn an M1 Mac with 16 GB of memory: usable with Granite 4.2 8B
Review your code before you commit
A local reviewer that reads each diff and flags logic, error-handling and security problems, without sending your code to a provider.
Uses: Ollama · Terminal
Open the guideModel Finder pick, estimated
- No GPU UsableOn a PC with 16 GB of RAM and no graphics card: usable with Granite 4.2 3B
- 8 GB GPU SmoothOn a PC with an 8 GB graphics card: smooth with Granite 4.2 3B
- 24 GB GPU SmoothOn a PC with a 24 GB graphics card: smooth with Qwen3-Coder 30B-A3B (MoE)
- Mac UsableOn an M1 Mac with 16 GB of memory: usable with Granite 4.2 8B
An offline voice assistant
Talk to a model and hear it answer, with speech recognition, the model and the voice all running on your computer.
Uses: faster-whisper · Ollama · Piper · Terminal
The Model Finder does not plan this job yet. The guide says what hardware it needs.
Open the guideDraft and reply to emails offline
Write, shorten and change the tone of emails with a model that never sees the internet.
Uses: LM Studio · Ollama · Desktop app
Open the guideModel Finder pick, estimated
- No GPU UsableOn a PC with 16 GB of RAM and no graphics card: usable with Qwen 3.5 4B
- 8 GB GPU SmoothOn a PC with an 8 GB graphics card: smooth with Qwen 3.5 4B
- 24 GB GPU SmoothOn a PC with a 24 GB graphics card: smooth with Gemma 4 26B-A4B
- Mac UsableOn an M1 Mac with 16 GB of memory: usable with Qwen 3.5 9B
Private AI for your business
Plan a self-hosted AI setup for a team: architecture, security, licences and the GDPR questions to answer first.
Home server, API
The Model Finder does not plan this job yet. The guide says what hardware it needs.
Open the guideA private tutor for learning
Use a local model as a patient tutor: it asks questions back, explains simply and writes practice quizzes.
Desktop app
Open the guideModel Finder pick, estimated
- No GPU UsableOn a PC with 16 GB of RAM and no graphics card: usable with Qwen 3.5 4B
- 8 GB GPU SmoothOn a PC with an 8 GB graphics card: smooth with Qwen 3.5 4B
- 24 GB GPU SmoothOn a PC with a 24 GB graphics card: smooth with Gemma 4 26B-A4B
- Mac UsableOn an M1 Mac with 16 GB of memory: usable with Qwen 3.5 9B
AI on your phone, offline
Run a small model directly on an iPhone or Android phone, with no connection needed.
Uses: PocketPal AI · MLC Chat · Phone
The Model Finder does not plan this job yet. The guide says what hardware it needs.
Open the guideA fully offline, air-gapped setup
Download everything once, move it across, and run models on a machine that never connects to the internet.
Uses: Ollama · Terminal
The Model Finder does not plan this job yet. The guide says what hardware it needs.
Open the guidePull structured data out of text
Get valid JSON back from a local model every time, ready to feed into a script or an app.
Uses: Ollama · llama.cpp · API
The Model Finder does not plan this job yet. The guide says what hardware it needs.
Open the guide
Not sure what your computer can handle?
The Model Finder asks what you want to do and what computer you have, then gives you one plan: the free app to install, the model to download and what speed to expect.
How we check these
Finder picks last verified 2026-10-06.
Every card links to a guide on this site. The model on a card is the Model Finder's pick for that job, and the Finder's picks pass a build check against our memory and speed model. A badge is an estimate for one modest machine of each kind, named on the card, not a measurement. Step-by-step recipes list the official documentation their steps were checked against, with the date. We only say a recipe was tried on real hardware when it was.
Questions
What can I do with a local LLM?
Most everyday text jobs: drafting and replying to emails, asking questions about your own documents, coding help in your editor, tutoring, and pulling structured data out of text. Add speech recognition and a voice and it becomes an offline voice assistant. Each card on this page links to a step-by-step guide for one of those jobs.
Do I need a graphics card?
No. Every card that shows badges has a plan for a PC with no graphics card. It will be slower than on a graphics card or a Mac, and the Model Finder picks a smaller model for it. The badges on each card say what to expect on each kind of machine.
Is running AI locally really private?
The model runs on your computer, so your prompts and files are not sent to an AI provider. The exceptions are the ones you choose: downloading the app and the model the first time, picking a cloud model or adding a cloud API key, and any add-on that calls an online service. Before you rely on it for sensitive work, check whether any part of your setup connects to the internet.
Which app should a beginner start with?
LM Studio if you would rather not use a terminal, Ollama if you are comfortable typing a command. Both are free and run the same open models. The Model Finder picks between them for you from your answers.