A small piece of an honest market for machine intelligence.

Most of today's machine intelligence is produced behind closed doors. A handful of labs control the data, the weights, the evaluations, and the distribution. The work is impressive, and the arrangement is fragile — a single policy decision, outage, or terms-of-service change can rewrite the surface that the rest of the industry stands on.

Bittensor proposes a different shape. Instead of a single platform, it's a permissionless network of subnets: each one a small economy where miners compete to produce outputs against a specific task, validators score the work, and TAO emissions flow to whoever actually moves the needle. The protocol doesn't pick a winner; it picks the criteria, lets the market run, and pays the result.

That changes who gets to participate in building intelligence. You don't need a billion-dollar cluster — you need to be useful against the subnet's scoring function. It's slower, messier, and more interesting than a closed lab. We think it's the right shape for the next decade.

Tensorbox is a Bittensor subnet — and the website you're reading is the operator interface around it. The subnet itself focuses on [subnet thesis — TBD]: the specific task we ask miners to optimize, and the scoring function validators use to grade them. Both will be documented in the docs once they're locked in.

Around the subnet we're building three things:

• Operator portal. A logged-in dashboard where miners and validators can see their own performance, scoring breakdowns, and reward history.
• Public telemetry. Open subnet metrics that anyone can inspect — no API key, no gate.
• Documentation we'd actually use. Specs, examples, and onboarding paths for both newcomers to Bittensor and existing operators evaluating our subnet.