# Take back your compute

AI got centralized fast. A handful of companies decide what a model can say,
what it costs, who gets rate-limited, and whose data trains the next
version. Most people who use AI every day have never run a model themselves,
never seen what's under the hood, and have no path to changing any of that —
just a login screen and a bill.

TokenBroker exists because that's not the only way this has to work.

## The idea is simple

Every laptop, desktop, and phone sitting mostly idle is a small amount of
real AI compute going to waste. Multiply that by a community instead of a
data center, and you get a network nobody owns outright — not one company,
not one server rack, not one point of failure.

- **You bring the hardware, or you bring a key.** A GPU that's idle most of
  the day. A phone that spends 20 hours a night doing nothing. A free-tier
  API key from a provider you already have access to. Any of these turns
  into real capacity for the network the moment you connect it.
- **The network pays you back.** Every real job your device or your key
  completes earns you credits — spendable on the exact same service you're
  contributing to. This isn't charity; it's a community that funds itself.
- **No single company can quietly change the deal.** The network already
  runs on more than one upstream (Ollama Cloud, Google AI Studio, and more
  as they're added) specifically so no single provider's rate limit or
  pricing change can take the whole thing down. Add your own key and you've
  made that redundancy a little bit stronger for everyone, not just
  yourself.
- **Nothing here is hidden from you.** The worker software is a visible tray
  app, not a background process you have to trust blindly. The live
  [network status page](https://tokenbroker.hopto.org/network) shows the
  network's real numbers to anyone, member or not, including the honest
  caveats about which figures are measured and which are estimated.

## Why it's worth doing

Centralized AI isn't just expensive — it's a single point of control. One
company's policy change, price hike, outage, or decision about what you're
allowed to ask becomes everyone's problem at once, because everyone depends
on the same door. A community-run network doesn't remove every tradeoff, but
it removes that one: capacity that no single party can unilaterally take
away, throttle, or price out of reach, because it was never any single
party's to begin with.

Putting real compute in real people's hands — instead of exclusively in a
few companies' data centers — is the whole point. Every new contributor,
every second provider key, every phone that joins overnight makes that a
little more true.

## How to actually do it

You don't have to choose between using the network and contributing to it —
most members do both.

1. **[Get started](getting-started.md)** — sign up, get an API key, make
   your first request. Takes minutes.
2. **[Join the compute network](compute-network.md)** — download the
   worker, or pair your phone, or add a free-tier provider key. Whichever
   you have, it counts.
3. **Tell someone.** A network like this only gets stronger with more
   independent contributors — not more centralization, more *independence*.
   Every person who joins with their own hardware or their own key is
   capacity nobody can take back. Referring someone who becomes a real
   active member gets **you both $5** — see
   [pricing & credits](pricing-and-credits.md#referrals-5-for-you-5-for-them).

Read the **[full system breakdown](system-breakdown.md)** if you want to
understand exactly how the pieces fit together before you commit, or
**[Tested & verified](tested-and-verified.md)** if you'd rather see the real
measurements and real bugs-we-found-and-fixed than take any of this on
faith — that's the point of documenting all of it in the open.

---

*Compute concentrated in a few hands is a choice, not a law of nature. This
is the other option.*