The essentials
- Assistants by area, with tools enabled one at a time
- WhatsApp, web chat, Slack, Teams, and Telegram
- Memory per person, sustained across channels
- Metrics with targets and proactive alerts
- Twenty-one prebuilt assistants by role, ready to adapt
A dashboard shows what someone decided to measure
A dashboard answers the questions someone anticipated when they built it. The new question —“which fields left a worse margin than last year?”— is a request to whoever knows how to build the query, and it takes days.
And nobody watches a dashboard all day: when a metric goes out of range, the dashboard shows it, but it doesn't alert.
How it decides who responds
A message arrives, the system identifies the person, resolves the conversation and chooses which assistant answers based on the topic and each one's declared specialty. If someone is named directly, that one responds.
A person is the same for the system even if today they ask via WhatsApp and tomorrow via the web chat: this is what allows an assistant to remember them across channels.
Swipe to see the full diagram →
What an assistant can do
An assistant is not just text: it has tools that an administrator enables one at a time — query your data, build a chart, read a spreadsheet or a PDF, schedule an alert, send an email, view the vigor of a field from satellite —. An agent with three tools and another with twenty coexist in the same company.
There is no need to start from scratch: there are twenty-one prebuilt assistants by role — grain marketer, grain elevator manager, livestock advisor, logistics coordinator, foreign trade advisor, among others — that adapt to each company.
Never guesses table names
This is what separates the platform from an assistant that improvises. An explorer walks through your connections and builds a semantic catalog: what tables exist, what columns, of what type, what they relate to, and what they mean in the business.
The assistant queries that catalog in natural language before writing a single line of SQL. If the data isn't there, it says so.
From conversation to alert
The leap that matters is going from “the assistant answers me” to “the system notifies me.” The rule is simple: anything that reaches a table can become a metric with its query and its target, and any metric can trigger an alert when it moves.
The situation room shows those metrics, and it is not a prebuilt dashboard: no two companies in the industry measure the same things. The producer looks at yield per hectare, the grain elevator at cell occupancy, the exporter at shipment compliance.
The color of the trend does not depend on whether the number goes up either: it depends on whether that is good for that metric. And if one has stopped receiving readings, it is flagged — an old number shown as if it were today's is worse than none, because decisions are made based on it.
Alerts go out through the channel where each person works, and by email when there is no other.
The limits, set on purpose
Outward-facing actions —launching a drone, opening a work order, notifying a supplier— come from a registered list, not from a free-form address. If the assistant could call any URL, any control placed on top of it would be bypassed by asking it to call something else, and it would also reach the server's internal network.
The assistant selects from that list and only fills in the declared data: the address, the key, and the message format never pass through it.
Combines with
iAgro Extract
Your data already exists. It just isn't arriving.
More about iAgro Extract →Data Platform
From raw data to an asset you can query.
More about Data Platform →Catalog
Everything you subscribe to, in one place.
More about Catalog →
