How Tendril works, end to end.
The honest, plain-language walkthrough of exactly what your device does, what you get back, and why you can trust it. Open a tab or tap your phone, help power open-source AI, and earn TEND Points.
Open a Tab: what your computer actually does.
Your browser runs real AI work in a sandbox, using only spare capacity. You can see it, cap it, and stop it anytime.
{{ s.body }}
Flip it. Spare capacity steps aside the moment you need your machine.
It is sandboxed, like any website. It never touches your files, your accounts, or your personal data. You can cap how much it uses, or close the tab anytime.
Tap to Train: how one tap teaches AI.
Every tap is a human-preference judgment. A few seconds of your taste is exactly the signal big labs lack.
Seconds per round, streaks, and levels, so people keep going. Ties post immediately. It respects your time.
A comparative judgment between two outputs, not editing the model yourself. Every tap makes the AI a little smarter.
Where it all goes.
Compute and judgment both feed one living, open dataset of which AI is best, in which language, at what. That is the actual product.
Model rankings, drawn from real inference running across millions of open tabs.
Human preferences, one tap at a time, the signal big labs cannot buy at scale.
Performance per language, so the world can see who is best in Swahili, not just English.
What you earn, and what it stands for.
TEND Points are your soulbound proof you helped build open-source AI. Recognition and rank that stay with you, and your on-ramp to what comes next.
Compute time on the open-a-tab path, and judged rounds on the phone. Tap-to-train work is weighted higher, because human judgment is scarcer.
Contribute, get verified, points bank to your handle. A token ($TEND) is on the roadmap, and your off-chain points are the system of record until then. A clean migration path, your record forever. You can merge a wallet later, but it is never required.
Why a Tendril rank actually means something.
The network only pays for work it can verify. Bots and farmers earn nothing, which is what makes the data sellable and your rank credible.
Multiple people judge each pair. Consensus is the signal. A lone vote against the crowd does not move the data.
Suspicious activity is throttled. Tap too fast or too perfectly and the network quietly stops counting it.
Known-answer tasks are seeded in to catch bots. Miss the ones with a clear right answer and your work is flagged.
Sandboxed, no files, no accounts, no personal data, pause or close anytime. Buyers can verify the quality, which is exactly why the open data is worth paying for.
One network, built by everyone, for everyone.
Today's AI is locked in a few warehouses and barely speaks most of the world's languages. Tendril fixes both, by being built by everyone.
Ask a top model a simple Swahili health question and it often answers in stiff, half-translated phrasing that a real speaker would never use. Human judgment is how that gets fixed.
"Kunywa maji mengi na pumzika." (Literal, flat, misses the warmth and context a real speaker expects.)
Underserved languages today's AI gets wrong, and the people fixing them:
This is running right now.
Everything above is live. Every number here is moving while you read this. The only thing missing is you.