By Web Asset House · data reviewed 2026-10-22
Local inference means the model’s inference computation runs on hardware you control. It can allow prompts and outputs to remain on that machine. That benefit depends on the application, model selection, tools, telemetry, sync settings, and network behavior. A product using the word “local” is not by itself proof that no data leaves the device.
Check the complete data path
- Confirm that the chosen model runs locally rather than through a cloud endpoint.
- Review extensions, web search, retrieval connectors, crash reporting, and cloud backups.
- Decide where chat history and uploaded documents are stored.
- Use operating-system permissions and network controls appropriate to your data.
- Test the actual workflow with a nonsensitive prompt while observing connections.
Downloads and updates are separate from inference
Fetching a model or updating an application requires external connections. A local inference workflow may still make those connections at other times. Ollama’s documentation distinguishes local behavior from optional cloud features; check the current settings and model choice for the software you use.
What this website does
The calculator uses static model and hardware records and computes estimates in your browser. It does not run an AI model, accept prompts, or upload your calculator choices to a third-party inference service. Following source and product links leaves this site and is governed by the destination’s practices. A future hosting provider may process ordinary request logs.
Open weights are not a universal permission
Access to model weights does not establish unrestricted use or a privacy guarantee. Check the official license, any use restrictions, and the software handling your data. Model-card metadata is informative but may be incomplete.
Primary sources
- Ollama: memory, context, local and cloud behavior
- llama.cpp: supported build backends
- Hugging Face: GGUF documentation
Questions, answered.
Does local AI guarantee that data never leaves the computer?
No. Inference location is only one part of the workflow. Tools, telemetry, cloud features, sync, and extensions may send data elsewhere.
Does this calculator send my prompts to a model?
No. It has no prompt input or inference service. It calculates hardware estimates from the static dataset in your browser.