ChatLima is the conversation inside Lima Juno — Agentic AI. A planner, a buyer or a director types a question the way they would say it to a colleague, and the answer comes back from live data in the systems you already run.
On-time in-full was 94.1% last week across 2,638 deliveries, 1.6 points above the week before. Two customers carry most of the gap.
Five steps, about four seconds, and every step visible afterwards.
A question arrives as a sentence. ChatLima first works out the intent: what the person wants to know, about which business objects, over which period. Materials, plants, suppliers, orders and deliveries are objects in the supply chain ontology: Lima’s built-in map of supply chain things and how they relate.
The ontology also says which registered tables hold those objects, because your team confirmed that mapping when the data was registered. So the step from “customer” to the right column in the right table is a look-up, and the answer rests on your confirmed meaning.
ChatLima then writes the query. When the objects live in one system, it is one query. When they live in three, the federated query engine splits the work, reads each source in place and joins the results. The rows stay in the systems that own them.
The query runs with the access rights of the person who asked. The result is presented the way the question implied: a table for a list, a chart for a trend, a sentence or two for a single figure. Each part of the working is shown beside the answer.
A reader who wants to trust a number finds its evidence on the same screen as the number.
A table, a chart or a short narrative, chosen to match the way the question was asked.
The ontology terms recognised in the question, so the meaning can be checked at a glance.
Each source that took part, named on the answer: your ERP, the planning suite, the warehouse system.
Seconds from the question to the answer, so slow sources show up early.
One click opens the query. An analyst can read it, copy it and run it themselves.
The plants, customers and columns the asking person is cleared for, stated on the answer.
These six parts appear on every answer, in the ChatLima screen and in the other places ChatLima works. Product.
One shared meaning answers the planner, the buyer, the director, finance, the steward and the data team.
The same conversation serves every chair, because the meaning underneath is shared. Each person gets the answer their access allows, from the same registered tables and the same confirmed mapping.
A shared, confirmed meaning is what turns a plausible answer into a correct one.
Most products that answer questions over data ask you to build a semantic model first: someone writes down by hand what each table and column means, which measures exist and how tables join. Juno supplies that meaning as a supply chain ontology and maps your columns onto it, with your team confirming each proposal.
That order matters for accuracy. When “delivery date” means the requested date in one system, the promised date in a second and the confirmed ship date in a third, a question about late deliveries has three possible answers. The confirmed mapping settles which one the company means, once, for every asker.
Gartner, 17 June 2025: organisations that give attention to semantics in AI-ready data can lift generative AI accuracy by up to 80% by 2027.
A published benchmark from 2023 ran the same questions twice over one insurance data set: 16% of answers were correct from the raw tables, and 54% from a knowledge-graph representation of the same data.
Juno also shows what it understood. When the wording allows two readings, ChatLima asks one clarifying question; when the evidence is thin it says so.
Showcase estate.
A. Novak registered 23 tables from five systems on Monday and the team confirmed the mapping on Tuesday. On Wednesday morning L. Tan asked ChatLima the question she had been answering with three exports and a spreadsheet: which open purchase order lines can still be cancelled this week, and what they are worth.
The answer came back in 4.2 seconds: 31 open lines worth $348,000 at plants 1210 and 1220, inside their cancellation window until Friday 18 September, grouped by supplier, with the query one click away. In the seven days that followed, 11 people asked 847 questions on the same registered tables.
The five steps in detail: intent and objects, the goal, the mapped tables, the query across one source or several, and how each question is classified first.
The spreadsheet route gives an answer that is hours old and privately defined.
The usual route to the same answer runs through the systems one at a time: open orders exported from one ERP, open orders from the second, the current plan from the planning suite, then a spreadsheet where material numbers are matched by hand.
The work takes hours and the answer ages while it is assembled. Each person repeats it with their own rules, so two planners can produce two versions of one figure. With ChatLima the matching happens once, in the confirmed mapping.
Any source registered through the data fabric: SAP, Oracle, planning suites, warehouse and logistics systems, lakes, cloud warehouses, databases, files and services. One question can read several of them and join the results.
Yes. Every answer runs with the asking person’s row and column access. Where a column is restricted, ChatLima states that it is restricted and names the role that can see it.
The showcase estate averages 4.2 seconds across three systems read in place. The time depends on the sources, the size of the result and the work the query does, and it is shown on every answer.
ChatLima asks one clarifying question, such as which date basis to use for “late”. Once you answer, the choice is kept for the rest of the thread.
The rows stay in your systems and Juno reads them there. Where history at scale is needed, Quanta lands a governed copy in your lake and ChatLima reads it there under the same access rules.
Each one builds on the one before it.
Reads your rows in place.
Connect and move ↗One record per real thing.
Master and govern ↗Apps for every decision.
Operate and decide ↗Asks and acts, at your level.
Ask and act ↗Bring two systems and one decision your planners make every week. In the first week we register the data, confirm the mapping with your team, answer your questions in ChatLima and run the first agent at the level you choose.
