Axis Agents
← Blog

16 November 2026 · Build story

How we built Reeve, a farm agent that names its sources

Reeve is an AI agent for farms. It keeps a farm's records, matches them to the rules, prices, pest bulletins and forecast around each field, and answers the farmer on Telegram by voice or text. It is a working demo on a sample farm, with France's rules only so far. The case study has the screenshots and a video. This post covers how we built it.

Mélanie Nolleau, Usama Maqsood and Najah Pokkiri co-created Reeve. Axis Spatial provides the spatial grounding.

The job was reading, not finding

We start every agent by mapping how the work moves today: who does what, with which information, and where it stalls. On a farm the information already exists. The farm has its own records and machine logs. The public sources are the parcel register, this year's CAP and aid rules, regional pest bulletins, prices and the forecast.

Each source lives in its own place and each one changes. Nobody has time to read them against each other every week. So the job was to do that reading: keep the records, match each public source to the fields it covers, notice what changed and remember what the farmer has already been told.

Match the source to the field first

A pest bulletin covers a whole region. The farmer has a barley field. A CAP payment depends on what grew on which parcel. If the agent answers from the regional bulletin, the answer is about the region.

So Reeve matches each source to the right field before it writes anything. The place, the parcel and the date stay attached to every piece of data. That is the spatial grounding Axis Spatial provides.

Every figure is checked against a source

Once the sources are matched, an agent works out the answer with tools that look up the farm's records, the rules, the bulletins and the forecast. Before the answer is sent, we check each figure against a source. A figure with no source is dropped, or Reeve says what it could not find.

The answers show this. A reply on next season's nitrogen cost lists each figure, says what Reeve does not have and names the sources it used. An answer on aid lists the checks it ran. The farmer sees where a number came from.

It works where the farmer already is

A farmer is rarely at a desk, so the main channel is Telegram. You ask by voice or text, in any language, and there is nothing to install. A short answer comes back in the chat. A Mini App inside Telegram shows the full answer with the figures, the sources and a map of the fields involved. There is also a web app.

Reeve speaks first, too. It raises a filing deadline, a field worth checking or a machine alarm without being asked.

Three questions it answers

  • Is there any aid I haven't claimed? Reeve reads the fuel and fertiliser logs against this year's aid rules. It lists what the farm could still file, where and by when.
  • Any pests to watch? Reeve compares years of regional pest bulletins with the farm's crops and sowing dates. It says which fields to check first. Whether to treat stays the farmer's decision, and only above the threshold.
  • Which days can I spray? Reeve puts the forecast, the work already booked and the spraying limits side by side and picks the best free day.

What the demo is built on

The farm's records are made up. The fields, rules, prices, pest bulletins and weather are real public data. The rules are France's. Every EU country has a parcel register and CAP rules, so the same approach should carry over. The next step is a pilot with one farm cooperative on real records.

If you run a cooperative or advise farmers, or you have a job like this in another field, book a call and tell us how the work is done today.