Prism is an AI workspace that refracts every task across the right place to run it — your own device, an attested cloud, or the community grid — and hands you a ledger line for each token. You set the privacy floor. Prism finds the fastest, cheapest path above it.
Type a task or pick a preset to see how Prism would split it.
A beam is a place your task can run. Prism scores each one on privacy, latency and price, then picks — or you pin one. Every beam returns the same shape of answer and the same ledger line.
Compact models execute on your own GPU through WebGPU. No network call, no account, works on a plane. Ideal for drafts, rewrites, personal notes and anything you'd never paste into a cloud.
Large open models run in confidential-computing VMs whose measurement Prism checks before every request. The receipt embeds the attestation so you can re-verify it later.
Thousands of contributors run mid-size models on idle hardware. Prism verifies a sample of answers, ranks nodes, and routes bulk work here when your privacy floor allows it.
Send one prompt to two anonymous models, pick the answer you prefer, and reveal who wrote it. Your votes feed a live Elo board that Prism's router reads when it chooses a model for you.
Enter the Arena →Imagine throwing a ball at a wall and, once in a very long while, it shows up on the other side without breaking anything. Tiny particles do exactly that, because they behave a bit like waves that can leak through thin barriers. It sounds like magic, but it's how some of your electronics work.
Quantum tunnelling is when a very small particle passes through a barrier it shouldn't have enough energy to cross. This happens because particles are also waves, and waves can spread a little way into walls. If the wall is thin enough, the particle can appear on the far side.
Give an agent a goal, a privacy floor and a dollar cap. It plans, spreads steps across the beams, and stops the instant the cap is reached — no surprise bills, no silent data leaks. Every step lands in your ledger.
Launch an agent →Notes, documents and past answers are encrypted in your browser with a passphrase only you know. When a task needs context, Prism decrypts just the relevant slices on the beam you allow — never on the grid, unless you say so. Share a slice with a colleague by link, with an expiry.
Open the Vault →The grid beam is powered by people. Open the Grid page, choose the hours your machine is free, and Prism routes verifiable batch work to you. No token to buy, no daemon to install — payouts settle weekly to the wallet or bank you connect.
One in eight grid tasks is silently duplicated to an attested node. If your answer diverges, it is rejected and your tier score drops; consistent nodes climb tiers and receive more work.
Every task, agent step and vault slice carries a level. The router only considers beams at or above it.
Every plan includes the Device beam for free, forever. Paid beams are metered per token at the price shown in your ledger before you send.
The whole Studio on your own hardware.
Attested and grid beams, agents, sharing.
For teams that need policy and audit.
Not a bill at the end of the month — a line per request, written before the answer streams. Beam, model, node, attestation, tokens, price. Filter it, export it, or pipe it to your finance stack by webhook.
It contains one, but the product is the router. The Studio is a workspace where tasks, agents and your vault share a privacy floor; the chat surface is just the quickest way to hand a task to the router.
Before sending to a cloud beam, Prism requests a hardware report from the VM, checks its measurement against a published allow-list, and refuses if it differs. The report hash goes in your ledger so you can check it again later.
Yes. Set your floor to L4 and every beam except Device disappears from the router. Agents inherit the floor and will refuse steps they cannot run locally.
Per accepted answer, in your local currency, settled weekly to a connected payout method. Duplicate-checks decide acceptance; tiers decide the rate.
No. Prompts on the Device beam never reach us. Prompts on attested beams are processed in memory and discarded. Grid prompts are visible to the operator that served them, which is why the trust ladder exists.
Yes. Attach a provider key and that provider appears as an extra beam with its own privacy level, metered at the provider's price with no Prism markup.
The Device beam is free and needs no account. Everything else is a passkey away.