Tools for running LLMs locally

Every tool on this site in one place. They run in your browser, send nothing anywhere, and use the same calculations as the rest of the site.

New tools

How much memory 32K, 64K or 128K tokens of context add, per model, KV type and runtime.

Which coding agent works with which local model on your hardware, with the context to set.

Paste one command’s output and get your GPU, VRAM and RAM filled in for the analyzer.

Pick your GPU or Mac and see every model that fits, with a speed range.

New to local AI? Pick a goal and a computer, get one plan.

Memory for one model at a chosen quant, context and batch, including LoRA and QLoRA.

Keep, add RAM, change GPU or rent: what to do to run a model you want.

The strongest card for coding, RAG, agents or vision at your budget.

A parts list sized for the models you want to run.

Break-even between a local GPU and hosted APIs.

Two or more models side by side: memory, speed and quality.

Measured and estimated tokens per second by GPU and model.

Build a /v1/systemone request and export curl, Python or JS.

Every figure comes from one compute module; the formulas are published on the methodology page.