Tendril FIRST BUILDS · ROUND 1 IS OPEN

Big AI labs don't build for causes that don't pay.
We do, with you.

Tendril is a network of everyday people who help power open-source AI from a browser tab or a phone. The Foundation points that work at the languages and causes the market skips, judged by the people who speak the language and live the problem.

Starts with language → Health answers Learning material Farming advice and more
Apply to build with us
We are early and have not announced partners yet, on purpose. We will not claim a partnership until it is real. Everything we build is open, provenance-stamped, and co-owned.
tendril.foundation
People from many communities building AI together
Tendril
Round 1
applications open
7,000
languages, ~100 served
The gap nobody is paid to fix

Language is the clearest case. It is not the only one.

The AI everyone uses was not built for everyone. If your language was not in the training data, the model fails it, and no big lab is rushing to fix that, because fixing it does not pay.

People in communities around the world building AI together
7,000
languages in the world. The major AI models are trained on roughly 100.
World Economic Forum 2025 (citing CDT)
2 wks
is roughly how often another language disappears from the world.
UNESCO, via WEF 2025
517
African languages benchmarked showed a pronounced gap, even major ones like Hausa and Wolof.
SAHARA benchmark, ACL 2025
The same gap, everywhere it does not pay
High human value, low commercial pull. The same gap opens across health, learning, farming, and more. Here is the machine that closes it.
Health Learning Farming + more
AI is only as good as the judgment behind it

The AI your community needs is missing one thing: data judged by the people who live there.

AI fails most of the world for one fixable reason: it lacks their data and was never judged by them. The fix is data judged and corrected by the people who live the problem, and that judged data is the one thing this network makes.

We cannot promise this fixes the world. We can promise the missing piece is something ordinary people can build, and we are starting to build it with the communities who ask.

Tendril makes
The local-language data and the evaluations.

Judged and corrected by the people who speak the language and live the problem. This is the one thing we own.

then
Open labs build
The clinic chatbot, the tutor, the farming app.

The labs and foundations that train open models do the rest, on that judged data. Not us, not yet built.

How it actually works

This is a real machine, not a concept.

One engine, judged by people who live the problem, feeding open models for many causes at once.

One contribution flows out to many causes
Open a tab
judge and correct answers
Tap a phone
run small models
PEOPLE
Tendril
The Tendril engine
Multiple people judge each item; we score agreement and run quality checks.
01
Human-preference judgments
02
Cleaned and translated language pairs
03
Evaluation sets
REAL OUTPUTS
Small-model data and human judgment. Not GPU training hours. Not a delivered tool.
How the help actually happens

Where Tendril ends and open labs begin.

AI fails these communities because it lacks their data and was never judged by them. That judged data is the one thing Tendril makes, not the clinic chatbot, the tutor, or the farming app.

The same engine points at any cause the market skips
What Tendril makes (solid, proven)
1

People judge

Local speakers rate and correct AI answers in their own language, from a tab or a phone.

2

Judged data & evals

Cleaned local-language pairs and evaluation sets that catch unsafe answers. The one thing we own.

What open labs do (dashed, not us)
3

Open models improve

Open-source AI labs and foundations train and test their own models on that data, on their own compute.

4

Real help, locally

A safer, more accurate model could reach people in their language. Potential, not delivered yet.

The one thing proven today is the sample above. The rest of this loop is how that work could travel, built by open labs on this data, not by us, and it does not exist yet for most of these causes.
One engine. Point it at any cause the market skips.

One root. Many tendrils. Language is where we start.

Every front below shares one trait: high public value, low commercial incentive, fixable with human judgment and small-model data, not frontier compute. Open a tendril to see how the help could work, and where Tendril stops.

SOLID = proven / Tendril DASHED = downstream, not yet
One root, many tendrils reaching many causes
Tendril
7,000 → ~100
Underserved languagesTHE ROOT · SAMPLE READY
7,000 languages exist; the major models train on roughly 100. The root cause we start from. Download a license-clean sample of the data format and provenance we produce.
Sample
How better AI could help

{{ r.help }}

Tendril makes the data, not the tool

{{ r.makes }}

TODAY, UNJUDGED
{{ r.today }}
judged data
IF LOCALLY JUDGED
{{ r.ifj }}
{{ r.disc }}
See all causes civic & legal · financial · mental health · elder care
We are early. We have not built for most of these yet, and we will never pretend we have. You tell us which cause is yours, and the first communities we work with decide what gets built first.
What you get as a partner

No money changes hands. You get agency, priority, and permanence.

All true on day one, before any track record exists. That is exactly why applying early is worth it.

1

You shape what we build first

The earliest partners set the roadmap: which language, which domain, which task the network runs next. Only true while we are early, which makes "no track record" the reason to apply now.

2

Open data and tools you co-own

Open datasets, eval sets, and tooling for your language or domain, openly licensed, with you credited as co-creator. Co-owned, never "donate your labor."

3

Priority access to a global network

Worldwide browser and mobile contributors producing judgments in your language and domain, a network you could never staff yourself. No GPU, no ML team, no budget on your side.

4

Founding collaborator recognition

Named as a founding collaborator if and when work ships for your cause, with your consent. Future-tense and consent-gated, because we will not claim it before it is real.

Who this is for

Underserved languages first. The same engine fits any cause the market skips.

If that is you, apply. The earliest partners shape what we build next.

What brings you here?
{{ sg.label }}
This is for you {{ segLine }} Apply →
How this is built and funded

We tell you exactly how this works. Hiding it would be the scandal.

Tendril Foundation is the public-good arm of the Tendril network. The network's commercial data work funds the open datasets we build for your cause, and that data stays open and co-owned by you.

Talk to a human, not an inbox bot
PUBLIC-GOOD ARM · THIS SITE
Tendril Foundation

Builds open datasets, evaluations, and tools for the causes the market skips.

COMMERCIAL · THE NETWORK
Tendril Labs

The contributor network. Its data work funds the open work. Same mission, different jobs.

What stays open and co-owned by you
The datasets The evaluations The tools
First Builds · Round 1 is open

We are choosing the first communities Tendril builds with.

A small team with one working slice can only build for a few communities first. The earliest ones shape what we build, get a direct line to the builders, and named-collaborator recognition if a build ships. Applications open in rounds, and Round 1 is open now.

Round 1
applications open now
You first
the earliest partners shape what we build
Apply to the First Builds cohort

Apply to build with us

A short application, reviewed in rounds by a small team. No funding forms, no money. Tell us who you serve and what you need built.

{{ segLabel }}change above
{{ tg.label }}
One to three sentences. This is the heart of the application.
Quick human check
{{ err }}
We use your details only to follow up about your application. No spam, no list-selling.

You have applied to Round 1.

We are a small team and we review personally, in rounds. We will be in touch. While you are here, download a license-clean sample of the data format and provenance we produce.

Download a sample dataset Back to the form
Tendril tendril.foundation

A public-good initiative building open-source AI data for the languages and causes the market skips. Part of the Tendril network.

Want to contribute as an individual? Join the Tendril network
© 2026 Tendril Foundation. A public-good initiative, part of the Tendril network. Open · provenance-stamped · co-owned