About Local AI
A practical, honest resource for running AI on your own hardware, from a laptop to a server rack.
Local AI covers the whole field of running models yourself: language models, coding assistants, agent frameworks, RAG systems, image generation, speech, and the hardware and tools that make it all work. It brings together things usually scattered across Reddit threads, GitHub READMEs, and Discord servers: news, guides, a catalogue of models and solutions, and two tools that answer the questions people actually arrive with, namely what you can run on your machine and whether running it locally is cheaper than an API.
We are built by people who think local AI matters, for privacy, for autonomy, and for the health of an open ecosystem. That belief shows in what we choose to cover. It does not show in how we report it. We are honest about where local AI falls short, and we are fair to cloud AI where it is the better answer. The way we write about all of this is set out in our editorial policy, which we treat as a hard requirement rather than a statement of good intentions.
Who writes this
Local AI is written and edited by Leon Mallett. Leon builds local-AI software, including Ragtag, a local-first desktop retrieval application, so the site is written from hands-on use of the models, tools and hardware it covers rather than from a distance. That he makes Ragtag is a relationship we disclose in full, and set out in the editorial policy; Ragtag is held to exactly the same standards as any other tool here, with no ranking or coverage advantage.
How we produce and check what we publish is meant to be visible, not taken on trust. Specifications, benchmark scores and licence terms are read at the primary source, attributed, and dated, so you can see where each figure came from and when; benchmark figures follow a documented policy on which analysis resources we rely on and why; catalogue entries carry the date they were last verified; and when we get something wrong we correct it in place with a dated note. Increasingly this includes our own hands-on testing of models rather than only summarising what others report.
Who it is for
People running a first model on a laptop, enthusiasts with dedicated GPUs, and organisations deploying inference for business, academic, or research use. Curiosity is the only prerequisite we recognise.