Miridia
Portal
Planning and AI

Miridia AI Studio

Ask questions about your operational data in plain English, and let the AI Autopilot watch orders, deliveries, and stock for you.

Miridia AI Studio has two parts. Ask Your Data answers a question about your data with a chart and a summary. The AI Autopilot runs agents on a schedule. The agents watch your operation and raise findings.

Ask Your Data is included in each plan. The AI Autopilot is an optional module.

Ask Your Data

Type a question in plain English. For example: "Which suppliers delivered late most often in the last six months?"

Miridia AI Studio then does these steps:

  1. It writes a query for your question.
  2. It runs the query against the data of your organisation only.
  3. It selects a chart and shows the result.
  4. It writes a short summary of the trend, the unusual values, and the exceptions.

Safe by design

  • Each query can only read data. A query cannot change or delete data.
  • Each query has a row limit.
  • Each query uses the credentials of the user who asks. It can only see the organisation of that user.

If a query fails, Miridia AI Studio reads the real schema and tries again.

Charts

Miridia AI Studio selects one of nine chart types:

ChartTypical question
KPI"What was the revenue last month?"
Bar"Which products sold the most?"
Line"How did orders change each week?"
Donut"What share of orders came from each channel?"
Stacked bar"Revenue by channel for each month"
Table"List the late purchase orders"
Gauge"How full is the warehouse?"
Heatmap"Orders by day and hour"
Treemap"Stock value by product group"

Each answer is a report card. Open the card to see the details. Miridia AI Studio keeps your conversation history.

Ask a precise question. Give the period, the location, and the measure. "Units of Madagascar Vanilla sold at Margate Storefront in August" gives a better answer than "vanilla sales".

AI Autopilot

The AI Autopilot runs agents that watch your operation. An agent can find a late shipment or an unusual number before a customer does.

Agent briefs

Tell an agent what to watch, in plain English. For example:

  • "Flag orders with an address that is not complete."
  • "Warn me when stock of a best seller goes below two weeks of cover."

Prebuilt templates are available:

  • Order data-quality sentinel: finds orders with bad data.
  • Stock watcher: finds low stock and stockout risk.
  • Revenue pulse: finds an unusual change in revenue.

Schedules

Each agent runs on its own interval, from every minute to once each day.

The findings inbox

An agent raises each issue as a finding. A finding has:

  • a severity: info, warning, or critical
  • a title and an explanation
  • a link to the record that has the problem

Open the findings from the notification bell. Acknowledge a finding when you act on it. Dismiss a finding that is not important.

Miridia AI Studio does not send the same issue many times. A recurring issue increases the occurrence count of one finding. A dismissed finding does not come back.

Transparent runs

Each agent run records a timeline. The timeline shows each check and the reason for it.

When an agent sees a pattern that repeats, it writes a simple fixed check for it. A later run that finds no problem uses the fixed check and does not call the AI model. This keeps the cost low.

Security

Each agent runs with its own credential. The credential can only read data, and only in your organisation. You can revoke the credential of an agent at any time.

Send findings to other channels

A finding can start a workflow. The workflow can send the finding to email, Slack, n8n, or another channel. Read Workflows.

AI models

Miridia AI Studio uses Anthropic Claude and DeepSeek models. You can select the model for each query and for each agent.

API reference