Chat
Ask anything in the browser. Pick any frontier model, type, and watch the answer stream in. Nothing to install, and conversations are never stored.
An equal shareof the frontier.For every human.
1,000,000 tokens* of frontier AI a day, free. No account, no card, no tiers.
Free.
No plan, no card, no trial that ends. Everyone verified in a region draws the same daily allowance, and nothing is for sale on top of it.
Daily.
1,000,000 tokens* refill at 00:00 UTC whether you spent yesterday's or not. Quests can add credits that never expire.
Yours.
One Self proof establishes that your passport is genuine and issued by that region's country. We keep a pseudonymous code that stops a second claim, and nothing else about you.
Your share.
One allowance, 12 model families. Switch between them mid-conversation without switching account, plan or key.
Use it in the browser, in the apps you already work in, or behind your own agents. Everything draws down the same daily budget, and shares idle for 7 days return to the pool for the next person waiting.
Use it wherever you already work
Three entry points, one budget. Every one of them draws on the same verified share.
Connect
Bring it into the apps you already use. One personal key works with any tool that lets you choose your own AI provider, and you can switch it off any time.
Agents
Give your assistants a free brain. Step-by-step guides ship for OpenClaw, Hermes and the Claude Agent SDK, so an agent can run on your share.
Every frontier family. One budget.
12 families today, one allowance. Switch model mid-conversation without switching account, plan or key.
Anthropic Claude
Industry-leading reasoning and long-form thinking.
OpenAI GPT
Flagship multimodal models from OpenAI.
xAI Grok
Real-time, opinionated, built by xAI.
Google Gemini
Massive context windows and vision from Google.
Meta Llama
The open-weights family most of the field is built on.
DeepSeek
Open-weights reasoning models that punch above their weight.
Mistral
Efficient European models that run well at small sizes.
Kimi
Long-context Moonshot models from China.
Qwen
Alibaba's open model family. Strong code and multilingual.
Zhipu GLM
The GLM family from Zhipu AI (Z.ai). Strong reasoning and code.
MiniMax
Long-context models from MiniMax, strong at agentic work.
NVIDIA Nemotron
NVIDIA's open Nemotron models, tuned for reasoning.
Three steps. About ninety seconds.
Prove you're a citizen
Scan your passport or national ID. A proof confirms it is genuine and issued by that region's country.
Pick a model
Switch between providers whenever you like. There are no tiers and no upsells, so the picker is a list rather than a pricing page.
Chat or build
Use the browser chat, or mint a key for your editor and your agents. Both draw on the same daily allowance.
Pick where you are.
Each region runs on its own subdomain with its own allowance and weekly intake. Verifying happens there, with that region's Self flow.
The things everyone asks.
That you hold a genuine government document issued by the country the region serves, and a code that stops one person claiming two shares. The check runs on your phone. Your name, date of birth, document number, photo and biometrics never reach us, and there is no account to log into afterwards.
Spots are finite, so a share idle for 7 days returns to the pool and the next person on the waitlist takes it. Use ubc now once inside that window and yours stays where it is.
No. Everyone in a region gets the same allowance, and there is nothing to upgrade to. That is the entire point: compute you cannot outbid someone else for.
Yes. Mint a personal key and use it in any app that lets you choose your own AI provider. The browser chat, your apps and your agents all draw on the same daily budget, and you can revoke a key at any time.
Requests are refused until 00:00 UTC, when the day's allowance refills. Nothing is charged, nothing is throttled silently, and quests can earn credits that sit on top of the daily pool.
Compute is a floor, not a luxury.
One proof, one share, no account. Pick your region and it takes about ninety seconds.