Practical guides to local AI
Run a model on your own computer, check what it says and turn document tasks into a workflow you can trust.
In this guide
Start with the task you need to complete. These guides take you from a first local model to checking a summary or extracting specific fields. Each one includes a practical starting point, common failure cases and links to the underlying documentation.
Choose your next step
Run local AI on Windows with Ollama
Install Ollama, try a local model and solve common setup problems.
Read the guide →Choose a local model for document summarization
Compare models on your documents, check factual accuracy and handle longer text.
Read the guide →Extract structured data with local AI
Define a schema, handle missing values and validate facts against the source.
Read the guide →How we check the examples
A convincing answer still needs to match the source. In the document guides, you can inspect fictional Swedish source texts alongside actual responses from local Qwen models. The surrounding explanations are in English; the original Swedish text and answers are preserved.
Those examples come from six short texts, three models from one family and two runs per text. They illustrate a checking method. They do not establish a general model ranking or predict accuracy on a full document collection.
We publish the test prompts, model identities and source material. The examples focus on content and instruction following. We do not publish a hardware profile or use these runs to compare response speed.
Updates to these guides
News covers a new model or tool when it appears. These guides collect the steps and checks that remain useful afterward. When a tested workflow changes, we update the relevant guide and explain the change here.
5 October 2026: First guide collection. Includes Windows setup, a model selection method for document summaries and a structured extraction workflow. The supporting Swedish examples use dataset version 1.0.