Launch a Local AI Agent Without Internet in 10 Minutes
Free model on your machine with agent features (tool calling, file handling) in LM Studio
- 1.Download and install LM Studio from the official website (Windows, Mac, or Linux). Run the installer - it takes 1-2 minutes.
The app is completely free. Choose the version for your OS. Mac automatically selects MLX for acceleration on Apple Silicon.
LM Studio ↗ - 2.Open LM Studio and go to the Discover tab (search icon). Find the Phi-4-mini Q4_K_M model - it takes ~3 GB of memory and works even on weak graphics cards.
Phi-4-mini is the best choice for a local agent with limited memory. It works with functions and tools right out of the box. The green rocket icon nearby means the model fits your hardware.
LM Studio ↗ - 3.Click Download and wait for the model to load (3-4 minutes on normal internet). The status bar shows progress in percentages.
The model downloads once. After that, it stays on your disk and works offline, without internet and without sending data anywhere.
LM Studio ↗ - 4.Go to the Chat tab, select Phi-4-mini from the dropdown list, and click Load Model. Wait for the model to load into memory (10-60 seconds).
When the model finishes loading, the input field at the bottom becomes active. This means the agent is ready to work.
LM Studio ↗ - 5.Open the Developer tab, find the Agent Mode toggle, and turn it on. Now the model can call tools and complete tasks step by step.
Agent Mode gives the local model the ability to call functions, work with files, execute code - all without the cloud and without API.
LM Studio ↗ - 6.Go back to Chat and write a task for the agent. Example: «Analyze this text. What is the main idea?» or «What functions can this agent call?»
Make sure Mode on the left side of the interface is set to Chat or Agent (visible at the top of the window). Press Enter or click Send.
LM Studio ↗PromptAnalyze this content and list the key points in a structured way. [Insert text or task description]
What this prompt doesThis is a basic prompt for your first test. Replace [Insert text...] with your own text or task. Phi-4-mini automatically breaks complex tasks into steps and executes them sequentially thanks to Agent Mode. - 7.Watch the response in the chat. The model shows the solution process step by step: which tools it chose, what data it processed. Everything works locally on your machine.
The agent's first response may take 30-90 seconds depending on task size and your hardware. That's normal. The next responses will be faster because the model is already in memory.
LM Studio ↗
- freeOllamaCompletely free, runs locally without internet after model download.
- freeDeepSeek R1 7B / Hermes 7B / Qwen 7BModels are distributed under open license, download is free, requires 5-7 GB of disk space.
- freePython + requestsPython and the requests library are completely free and open source.
Why today
AI services change fast - interfaces and free limits may differ from what's described.