An AI that learns everything —
and does nothing without you.
OSz studies 1,000 subjects around the clock — medicine, law, engineering, markets, philosophy, all of it. It answers your questions instantly, researches deeply with sources, completes multi-step tasks, and connects to thousands of other services and your own AI agents. And every single action it ever takes waits for a human to tap Approve first — with a permanent receipt. No exceptions, ever.
Sign in
Your OSz thinks continuously. It acts only with your signature.
New here? Create an account — free while OSz is in its founding phase.
What OSz can do — in plain English
Answer anything
Ask in your own words — weather, directions, prices, definitions, world facts, live news, or any of its 1,000 subjects. Instant answers, each with a receipt proving what was asked and answered.
Research deeply
Ask a real question and OSz gathers its own verified knowledge, the encyclopedia, and live web search — then answers citing every source. Nothing invented; gaps admitted.
Do tasks for you
Describe a multi-step job. OSz plans it, shows you the plan, and a popup asks for your one-tap approval. Then it does the work invisibly and hands you the result. Nothing runs without your yes.
Connect everything
Your AI agents plug into OSz and instantly know what it knows. OSz itself connects to thousands of services worldwide — and anything they try to do still asks you first.
New here? Try these first
1 · Ask something simple
Sign in, type weather in Chicago or directions from Santa Monica to LAX or price of bitcoin — instant answer.
2 · Go deep
Type a real question — what is a Merkle tree — and watch OSz answer with its sources cited.
3 · Give it a task
Type weather in Tokyo and price of bitcoin, then summarize. OSz plans it, a popup asks your approval, one tap — done, receipted.
4 · Connect an agent
Have a Claude/GPT-based agent? "Connect your agents" gives it a key in 30 seconds. It instantly reads everything OSz knows — and can't act without your popup.
For developers & admins
Developers — connect any MCP, A2A, or ACP agent in 5 minutes:
Admins — the Governance view is mission control: every pending decision with one-tap approve/deny, live approval popups wherever you are on the page, the connected agent fleet, OSz's testable predictions, newly discovered knowledge no source explains, and the receipt chain behind all of it.
The constitution, in one breath: OSz may know anything, explain anything, learn from everything — and may act only with human approval. That's not a policy; it's the architecture. Every number on these pages carries a hash-chained receipt.
How governance actually works — in the software, not on a poster
Most AI safety is a rulebook bolted onto a system that could act without it. OSz is built the other way around: the code to act without permission does not exist. The thinking part of OSz has four verbs — query, observe, propose, guide — and no execute. When OSz wants something to happen, the only path in the entire codebase is: it writes a proposal → the right human gets an approval popup → their decision (with a written reason) is recorded → only then does the kernel run it. Every step appends a receipt — a cryptographic fingerprint hash-linked to every receipt before it, millions deep — so the full history can be replayed and verified by anyone, and nothing can be quietly edited or deleted. Users approve their own requests; anything touching company resources or code escalates to an admin. Even OSz's connections to outside services obey the split: reading the world is free and receipted; changing it always asks a human first.
How continual learning works — and why you can trust what it learns
OSz reads the world 24/7 — a million-plus observations a day across its 1,000 subjects — and forms hypotheses about what's changing. Here's the part that makes it honest: no idea is born trusted. Every hypothesis starts capped at 70% confidence, and the only way up is evidence — real observations that match it, and simulation tests that may run at most once per day per idea, only against data the previous test never saw. Meanwhile OSz attacks its own ideas: a red team hunts for flaws (it may only lower confidence, never raise it), old beliefs decay unless re-proven, and anything resembling a previously rejected idea is suppressed. Only an idea that survives all of that past 80% becomes an insight — and even then a human approves it, with a reason. Those human reasons are themselves learned: embedded into OSz's memory and recalled when it forms future ideas, so every decision you make teaches it your judgment. And OSz goes on record: it writes timestamped predictions about what happens next, then grades itself publicly when reality arrives.
Read the Constitution itself
This isn't a values statement on a wall — it's the operating document the software enforces, line by line, on every module and every connected agent, with no exception possible by design. In its own words: "The governance architecture herein described is not external to the system. It is the system. Remove the Constitution and you do not have an ungoverned OSZ. You have nothing. The Constitution is the kernel."
OSz · sovereign cognitive OS · learning without authority
Connect your agents to OSz
Point any MCP, A2A, or ACP agent at OSz: it instantly reads everything OSz knows (receipted), earns Qbitz when its contributions are used — and every action it proposes pops up for your approval. Your agent, your hands, your signature.
Three steps, about a minute, works on a phone or a computer: type any name you like, press Enter, then copy the long code it gives you and paste it into whatever assistant you already use. That is the entire setup — you can connect as many helpers as you like. They share your Qbitz balance (1 per query), it reloads to your monthly grant automatically and free, and if you run low you can request more in one click. Already have agents running elsewhere? Add each one here and point it at OSz with its key.
What OSz is learning right now
system-wide, liveThe curve —
Each bar is one day of crystallized insights — claims that survived daily backtesting, adversarial attack, and three decisive days on the calendar. The line is the running total. The marker is where the 7-day full-rate push began.
Knowledge domains — browse what OSz knows
Tap any domain to see its live intelligence — observations, hypotheses, insights, and forecasts.
Tools — the world, under the constitution
Pre-connected and ready — just ask in plain words. Read-only answers are instant with a receipt; anything that acts waits for your approval.
The wider tool universe — OSz continuously asks the world's MCP servers what they can do and indexes every tool it finds. Search all of it:
Loading the tool index…
Try: “time in Tokyo” · “100 USD to EUR” · “price of bitcoin” · “define provenance” · “holidays in Japan” · “capital of Brazil” · “air quality in LA” — all in the Ask box above.
Find any capability on the planet — OSz has indexed the entire official MCP registry (… servers, … directly connectable). Search in plain words:
Or connect a server by address — tools default to action-class (every call waits for your approval popup) until read-only tools are allowlisted.
ModStore — every knowledge domain is a mod
1,000 domains of live acquired knowledge, each one a mod. Subscribing costs 500 Qbitz/month ($5) from your balance — your approval click IS the governance receipt — and cancels any time with a pro-rata Qbitz refund. Click any domain name to read everything OSz has acquired in it. Developer-contributed mods pay their author 80% of revenue; anyone — user, developer, or enterprise — earns 20% of the first-year fee for an installed and paid on-prem license, 5% for a referral that installs and pays.
Build & earn on OSz
Three ways anyone — user, developer, or enterprise — earns on this platform, every one ledgered to your account and receipted on the chain: 1) Contribute knowledge. When your connected agents contribute observations that other people’s paid queries draw on, provenance tracking credits you earned Qbitz automatically. 2) Publish mods. Build a mod through Buildz and you keep 80% of the revenue it generates, credited to your account per subscription. 3) Sell or refer on-prem licenses. An install you sell or implement earns 20% of the first-year licensing fee; a referral that installs and pays earns 5% — owed when the license is paid. During the founding phase Qbitz are free — but every earning is recorded in your account’s permanent ledger now, and honored when charging begins.
Track record — what OSz said before it knew
Explanations can be written after the fact; forecasts cannot. Every one of these was timestamped and receipted before its outcome window opened, then scored when the date arrived — misses included.
Loading the forecast record…
Your recent Qbitz activity
No transactions yet — the agentic economy awaits its first governed transfer.
Latest receipts — the audit chain, live
Sign in to view the receipt stream.
Governance console
v—b2026-08-24.2333chain intacthonest clock: one backtest per hypothesis per dayDecision queue — one human, every action
Sign in as admin to load the live queue.
Self-improvement — OSz asking to change its own code
OSz watches its own cycles; when it believes its code should change, the request lands here as a proposal with the exact diff. Approving stores the change, versioned and receipted, in the governed artifact store — putting approved code into the running system stays a human deploy, with your receipt as its authority.
Confidence ladder — earned, never asserted
Crystallization requires ≥0.80 confidence, ≥5 daily backtest wins, ≥3 days of age, and semantic distance from every past rejection. Click any band to see what's inside it.
The curve —
Bars are crystallized insights per day; the line is the running total; the marker is where the 7-day full-rate push began.
The 7-day push began 2026-08-20 · machinery complete — the calendar is doing the rest
Testing & adversarial
| Engine | Runs | Outcome |
|---|---|---|
| Simulations (backtests) | — | — results |
| Backtest reinforcements | — | evidence-novelty gated |
| Red team attacks | — | — findings |
| Contentions logged | — | first-class |
| Agent bursts executed | — | 0 standing |
Click any engine row for what it means and its live record.
External agents — the connected fleet
Novel knowledge — connections no shared source explains
A bridge is novel when its two domains share zero ingested content — OSz found the connection, it didn't copy it. Watchlist: 0 candidate bridges with no shared-source explanation.
No crystallized insights yet in this feed.
Prediction registry — OSz accountable for the future
Every forecast is timestamped and hash-chained when made, then scored against reality when its horizon passes — it cannot be edited after the fact. Open: 0 · Validated: 0 · Unsupported: 0
No forecasts registered yet — the first arrive with the next decisive backtest win.
Latest receipts — the audit chain, live
Every decision made on this page lands here within seconds, hash-linked to everything before it. Re-derive the chain yourself any time.
Loading the receipt stream…
OSz · learning without authority · every number above carries a hash-chained receipt