More out of the same plot
- Name the pest on the leaf from a photo
- Work out rates for the plot you actually have
- Plan planting around this season, not last one
- Check what the crop is worth before you sell
Grow is Fantail AI's non-profit programme — practical AI in the hands of the farmers, teachers and health workers already doing the work. In their language, on the phone they already carry, free to use.
In the language of the place. On a phone the team already shares. Free to use.
a question→
in your language→
answered on the phone→
today
Farming, school and health are where a rural day actually goes. Improvements in each compound over a season — and all three are held up by the same thing: a question nobody nearby can answer today.
None of this is advice arriving from outside. It's a reference that finally answers back — the judgement stays with the person who knows the place.
In the language you actually speak, out loud if your hands are full, with a photo if that's quicker. The answer comes back in the same language, on the phone in your hand, and it's free to ask — the programme carries the cost, not the person with the question.
Straight answer on the technology: today the models are hosted, so asking a question needs a connection. Running them on the handset itself — no signal required — is where this is going, and it's the single biggest thing that would change how far it reaches. We'd rather say that plainly than let the page imply it already works that way.
What's missing is this season's answer to a question nobody within a day's travel can settle. What is on this leaf. What rate for a plot this size. What the crop is fetching this week. The knowledge already exists — it just never arrives while the decision is still open.
The knowledge nearly always exists somewhere. It just tends to arrive after the planting window shut, after the class moved on, after the child got worse. Timing is most of the value.
The gap isn't knowledge and it isn't skill. It's the distance between having a question and having someone to ask.
The whole deployment is a phone the team already shares. No hardware to procure, no device per person, no login each. Anything that starts with "first you'll need to buy" has already failed the test that matters.
Asked in the language of the place and answered in it — out loud, because most of this work happens standing up with your hands full.
No hardware to buy and no device per person. The handset the team already passes around is the entire deployment.
Not a milestone at the end of the plan — the thing the plan is for. Everything needed to rebuild it goes with it.
Every deployment starts with an organisation that has been doing the work for years and knows exactly where the hours go. They set the brief and keep the judgement. We do the model engineering — the part that's actually ours to do — and then it's theirs.
It's the word most easily said and least often meant, so here's the literal version: at the end, the organisation running it can change the tool, rebuild it, and keep it going without us.
Not a pilot that quietly ends with the funding round.
Not a licence, a seat count, or a local reseller agreement.
It's a tool the team could rebuild without us.
Two ways in, and they need each other — a deployment without the operating organisation driving it is the exact failure this programme exists to stop repeating.
Grow is the non-profit side of Fantail AI. It doesn't sell anything, it isn't a route into the commercial products, and partner organisations are never a sales pipeline. When the two sides share engineering, the work flows outward — into the open — not the other way round.
Ask us what's running, what's a prototype, and what's still only an idea. You'll get the honest list — a programme that oversells its stage does more damage than one that moves slowly.
Technology tends to arrive in rural places with conditions attached and somebody else's agenda behind it. These are our terms, stated up front so you can hold us to them.
No training on what people ask. Questions and answers are not used to train a model — ours or anyone else's. Not aggregated, not de-identified, not "for research".
No diagnosing, no decisions about people. On health it explains, translates and recalls what the guidelines say. It does not diagnose, score eligibility, or decide who gets seen first — the health worker does, as they already do.
No surveillance work. Not identification, not tracking, not profiling — for any government, donor or partner, at any price, however the request is framed.
No pilot-and-leave. If there's no plausible handover at the end, we don't start. A demo that dies with the funding round wastes everyone's time, and theirs is not less valuable than ours.
No English-only launches. If it only works in English it isn't finished, and it doesn't ship. Language coverage is scope, not polish.
No overstating what's built. You'll be told which parts are running, which are prototypes, and which are still just intentions — including the awkward ones.
The service that answers questions is a few hundred lines you can read end to end, and the instructions are written down properly. The point is to be copyable — if an organisation we've never spoken to rebuilds the whole thing and runs it themselves, the programme worked.
of what people ask are ever used to train a model, ours or anyone else's.
is the whole deployment — no hardware to buy, no device per person.
the app, the code and the instructions belong to the partner organisation.
and not a sales channel — Grow sells nothing and never passes partners to the commercial side.
If you run something rural — an extension service, a school network, a clinic — and you already know the twenty questions that eat your week, that's the conversation this page is for.