three ways sol thinks

three ways sol thinks

solstone gives you three ways for sol to think about your journal — you choose during setup, and you can change your mind anytime in the thinking app.

local — the default

on a capable machine, sol thinks right here: a bundled local model, set up in the background during install. capable means roughly 6–8 GB of GPU memory (linux, AMD/NVIDIA/Intel via Vulkan) or 16 GB on apple silicon — sol checks before anything downloads, and tells you plainly if your machine falls short rather than guessing. see what your machine needs to think locally for the full detail.

your thinking never leaves the machine on this lane.

your own AI engine (byo)

bring your own token — Claude, Gemini, or GPT — or point sol at any OpenAI-compatible endpoint you run yourself. your billing, your control; the key stays in your journal, and sol pbc is never in the path.

confidential processing — coming, scouts first

let sol think off your device, on confidential hardware sol pbc runs — without a third-party AI provider in the path. only the parts of your journal sol is thinking about go to the service, processed and not retained: no content kept, no human review, nothing used to train. your journal itself stays on your device either way. the journal checks the service's attestation before it sends anything, and won't send if it can't verify.

the hardware behind this is confidential GPU hardware sol pbc operates on Azure, verified by the attestation check above — see confidential processing on the services portal for the full disclosure.

not open yet — scouts get first access, and there's no pricing yet either.

switching

open the thinking app (solstone's web ui, by default http://localhost:5015). one lane is active at a time, switching is instant, and sol never silently picks one for you — if nothing's chosen, it says so rather than reaching for the cloud on its own.